H ighlights 

“The future is already here. It’s just unevenly distributed.”
 (William Gibson) The State of AI 2019: Divergence 

For entrepreneurs, executives, investors and policy-makers, The State of AI 2019 is an accessible, comprehensive guide to the reality of AI today, what’s to come and how to take advantage. Download at www.stateofai2019.com. Highlights: 1. The race for adoption
 Increasing adoption of AI masks a growing divergence among nations, across industries and between companies. 2. The war for talent
 While demand for AI professionals exceeds supply, winners and losers are emerging in the war for talent. 3. The advance of technology
 Emerging techniques (transfer learning, reinforcement learning and generative AI) are enabling new possibilities. 4. The Disruptors: Europe’s AI startups
 Our unique analysis of Europe’s 1,600 AI startups reveals a maturing ecosystem bringing creative destruction to new industries. 5. The implications of AI AI will reshape sector value chains and enable new business models, while risking greater inequality and the erosion of trust. 3 The State of AI 2019: Divergence 

MMC Ventures is a research-led venture capital firm. We’ve backed over 60 early-stage, high growth technology companies since 2000. AI is a core area of research and investment.
 We’ve made 20 investments into many of the UK’s
 most promising AI companies. If you’re an early stage AI company, get in touch
 to see how we can accelerate your journey. www.stateofai2019.com We’re MMC Ventures 4 MMC Ventures Research David Kelnar: Partner & Head of Research
 Asen Kostadinov, CFA: Research Manager Explore MMC’s cutting-edge research at mmcventures.com/research. 

AI: Coined in 1956, a general term referring to hardware or software that exhibit behaviour which appears intelligent. Machine learning: Modern AI, which enables software to perform difficult tasks more effectively by learning through training instead of following sets of rules. Deep learning: A form of machine learning that approximates the function of neurons in a brain to solve problems. New to AI? Read our primer in www.stateofai2019.com 5 Before we begin… 

The State of AI 2019: Divergence 1. The race for adoption 

AI may be the fastest paradigm shift in technology history. But increasing adoption masks a growing divergence between leaders and laggards – among nations, between industries and across companies. The race for adoption The State of AI 2019: Divergence 7 

The State of AI 2019: Divergence AI adoption is proliferating 8 One in seven large companies has deployed AI Source: Gartner, 2019 CIO Survey: CIOs Have Awoken to the Importance of AI, figure 1, 3 January 2019. Base: All answering, n = 2,882. What are your organisation’s plans in terms of artificial intelligence? One in seven large companies have deployed AI Enterprise plans to deploy AI No interest Plan to deploy in 2 to 3 years Will deploy in 12 to 24 months Will deploy in next 12 months Have already deployed 9% 29% 25% 23% 14% • Today, one in seven large companies has adopted AI. • In two years, two thirds of large companies will have live AI initiatives. 

The State of AI 2019: Divergence AI is ‘crossing the chasm’ 9 In 2019, AI ‘crosses the chasm’ to the early majority Source: Everett Rogers, Geoffrey Moore • By the end of 2019, over a third of enterprises will have deployed AI. • In 2019, AI ‘crosses the chasm’ to the early majority. • Adoption of AI has progressed rapidly from innovators and early adopters to the early majority. AI adoption is ‘crossing the chasm’ to the early majority Innovators Early adopters CHASM Early majority Late majority Laggards Population 2.5% 34% 34% 16% 13.5% 

Adoption has been enabled by: • the prior paradigm shift to cloud computing • the availability of plug-and-play AI services from global technology vendors • a thriving ecosystem of AI-led software suppliers. 10 AI: the fastest paradigm shift in enterprise tech history? The State of AI 2019: Divergence Over three years, the proportion of companies with AI initiatives will have grown from one in 25 to one in three (Gartner). 

• Adoption is being catalysed by companies’ growing conviction in AI’s potential. • A greater proportion of executives believe AI will be a ‘game changer’ than any other emerging technology – including cloud, mobile, IoT, blockchain or APIs. 11 Great expectations are fuelling adoption The State of AI 2019: Divergence 2019 CIO Agenda
 Which technology area do you expect will be a game-changer for your organisation. Top performers
 (n=230) Typical performers
 (n=2,329) Trailing performers
 (n=276) AI / Machine Learning 40% 25% 24% Cloud (including XaaS) 12% 10% 14% Mobile (incl. 5G) 7% 6% 5% Internet of Things 6% 10% 11% Blockchain 5% 4% 5% ERP 1% 3% 3% AI tops the list of technologies companies perceive as ‘game-changing’ Source: Gartner, 2019 CIO Survey: CIOs Have Awoken to the Importance of AI, figure 1, 3 January 2019. 

12 Use of AI applications is advancing across a broad front The State of AI 2019: Divergence • Today’s enterprises are using multiple types of experiential and analytical AI applications. • One in ten enterprises now uses ten or more AI applications (Gartner). The State of AI 2019: Divergence Chatbots have displaced fraud detection as the top use of AI in 2019 Does your organisation use any of these artificial intelligence (AI) based applications? 2019: n = 2,791; 2018: n = 2,672. Multiple responses allowed. Source: Gartner, 2019 CIO Survey: CIOs Have Awoken to the Importance of AI, figure 1, 3 January 2019. 

13 Popular use cases: asset management The State of AI 2019: Divergence • Investment strategy
 Improve a firm’s investment strategy by synthesising research and data, and incorporating broader data sets including unstructured information. • Portfolio construction
 Augment or automate an asset manager’s process of portfolio construction. • Risk management
 Improve risk management by incorporating broader data sets and enhanced analytical processing. • Client service
 Natural language AI enables client enrolment, support and self-service. Chatbots enable account managers to query client details and understand developments with client portfolios in seconds. 

14 Popular use cases: healthcare The State of AI 2019: Divergence • Diagnosis
 Replace complex, human-coded sets of probabilistic rules and identify subtle correlations between vast, multi-variate data sets to deliver scalable, automated diagnosis. • Drug discovery
 Synthesise information and offer hypotheses from the 10,000 research papers published daily; predict how compounds will behave from an earlier stage of the testing process; identify patients for clinical trials. • Monitoring
 Synthesise signals from inexpensive wearable devices worn by patients to deliver clinical-grade monitoring; enable large patient groups to be monitored in real-time by a single nurse. 

• Risk assessment
 Assess the risk of individual policies more accurately than rules-based systems by detecting
 non-linear patterns in multi-variate data sets and making more accurate projections. • Claims processing
 Automatically extract and classify structured and unstructured information from insurance policies and claim forms. • Fraud detection
 Identify fraudulent transactions, while reducing false positives, more effectively. • Customer service:
 Chatbots using natural language processing can offer 24/7 product information and answers to policyholders’ enquiries in a scalable, inexpensive and personalised channel. 15 Popular use cases: insurance The State of AI 2019: Divergence 

16 Popular use cases: law and compliance The State of AI 2019: Divergence • Case law, discovery and due diligence
 Natural language processing AI can identify, classify and utilise content from databases and unstructured documents at scale and speed, saving legal firms time and cost for document review. • Litigation strategy
 Analyse past judgements at greater speed, granularity and subtlety. By anticipating the probability of different outcomes, inform and enhance strategic decision-making. • Compliance
 Flag potential compliance breaches in real-time, before they occur, with sufficient accuracy to enable broad deployment. 

17 Popular use cases: manufacturing The State of AI 2019: Divergence • Predictive maintenance
 Identify subtle patterns in data from vibration, temperature, pressure and other sensors to identify leading indicators of equipment failure and save expensive, unplanned downtime. • Asset performance
 Improve the operation of high value assets, including gas and wind turbines, to optimise yield. •Utility optimisation
 Anticipate, and align, utility consumption with process requirements in real-time, lowering utility consumption by 5% or more. 

18 Popular use cases: retail The State of AI 2019: Divergence • Customer segmentation
 Access additional data sets (e.g. social media data) and undertake more granular analysis than rules-based systems allow, to optimise segmentation, channel selection and messaging. • Content personalisation
 Access additional unstructured data sets for analysis, and improved multivariate analysis to identify more subtle correlations than rules-based systems can detect. • Price optimisation
 Identify correlations within and between data sets, to better optimise for factors including price elasticity, revenue, profit, availability and phases in product’s lifecycle (introduction or end-of-life). 

19 Popular use cases: transport The State of AI 2019: Divergence • Autonomous vehicles
 Enable vehicles to sense and identify physical features and dynamics in their environment, from road lanes to pedestrians and traffic lights, with a high degree of accuracy. • Infrastructure and system optimisation
 Detect patterns and optimise complex data to address traffic, congestion and infrastructure challenges in transport systems. • Fleet management
 Optimise pick-ups, route planning and delivery scheduling to maximise asset utilisation, while considering economic, social and environmental impacts. •Control applications
 Address prediction and optimisation challenges presented by air traffic control, vehicle traffic signalling, and train control. 

20 Popular use cases: utilities The State of AI 2019: Divergence • Supply management
 Predict changes in supply, including those caused by the intermittency of renewable resources, enabling smaller reserves and cost savings. • Demand optimisation
 Identify detailed patterns in consumer behaviour to move consumption of energy from periods of peak use and high prices to times of lower demand and cost. • Security
 Identify abnormal patterns in network behaviour to detect breaches in network security that elude traditional programs. •Customer experience
 Chatbots, powered by deep learning algorithms, offer consumers self-service account administration, product information and customer service. 

‘Deployed AI’ (% of companies) Source: Gartner ‘Deploying AI in the next 12 months’ (% of companies) Source: Gartner 21 Globally, China leads the race in AI adoption • Twice as many enterprises in Asia have adopted AI, compared with companies in North America, due to: government engagement; a data advantage; and fewer legacy assets. • Chinese companies have a dual advantage: more permissive policies than Europe regarding use of personal data; and less siloed data within companies. Deployed AI 25% 1% Deploying AI in the next 12 months 30% 1% Rapid adoption of AI overall masks a growing divergence – among countries, across industries and between companies. 

Adoption of AI is uneven across, and within, sectors Source: Gartner 22 Sector adoption is in a state of flux The State of AI 2019: Divergence Divergence is also evident across industries: • ‘Early adopters’ (financial service and high-tech companies) invested early in AI, sustained their investment, and today maintain a lead. • ‘Movers’ (retail, healthcare and media) have awoken to AI’s potential and are rapidly catching up. • Government agencies, education companies and charities are laggards in AI adoption. Utilities Wholesale trade Government Education Charity Natural Resources Financial Services Health (Providers) Transport Manufacturing Media Health (Payers) Retail Telecoms Software & IT Services Insurance Adoption of AI is uneven across, and within, sectors Percentage of respondents that have deployed AI or plan to within 12 months 48% 46% 44% 44% 44% 42% 40% 39% 38% 37% 35% 33% 32% 26% 26% 25% Vulnerable members of society may be among the last to benefit from AI. 

Leaders and laggards face different adoption challenges 23 Leaders and laggards face different challenges to adoption Source: S. Ransbotham, P. Gerbert, M. Reeves, D. Kiron, and M. Spira, “Artificial Intelligence in Business Gets Real,” MIT Sloan Management Review and The Boston Consulting Group, September 2018. • Among companies, laggards are struggling to gain leadership support for AI and to
 define use cases. • Leaders’ challenges have shifted from ‘if’ to ‘how’. Leaders are seeking to overcome the difficulty of hiring talent and addressing cultural resistance to AI. 20% 40% 60% 80% 100% 0% Attracting, acquiring and developing the right AI talent Competing investment priorities Security concerns resulting from AI adoption Cultural resistance to AI approaches Limited or no general technology capabilities (e.g. analytics, data, IT) Lack of leadership support for AI initiatives Unclear or no business case for AI applications Passives Experimenters Investigators 2017 data What gets in the way of AI adoption? Pioneers Leaders and laggards face different challenges to adoption 

The State of AI 2019: Divergence There is a growing gulf in corporate understanding of AI 24 • Among AI laggards, fewer than two in ten understand the implications of AI. • Among leaders the reverse is true; eight in ten understand its dynamics. There is a gulf between leaders’ and laggards’ understanding of the implications of AI Source: “Reshaping Business With Artificial Intelligence”, MITSloan Management Review in collaboration with The Boston Consulting Group Leaders are extending their advantage in AI through better understanding, faster learning and greater investment. 

The smart are getting smarter Source: “Reshaping Business With Artificial Intelligence”, MITSloan Management Review in collaboration with The BCG. AI leaders are extending their advantage through greater investment Source:“Artificial Intelligence in Business Gets Real,” MIT Sloan Management Review and The BCG, September 2018. 25 Leaders are learning faster and investing more The State of AI 2019: Divergence • The smart are getting smarter. In the last year, between half and two thirds of AI leaders improved their understanding of AI to a great extent. Just a tenth to a fifth of laggards did so. • Leaders are extending their advantage through greater investment. Nine in ten AI pioneers have increased their investment in AI in the past year compared with one in five passives. 10% 20% 30% 40% 50% 60% 70% 0% Passives Experimenters Investigators Pioneers Fig. 4b: How much are organisations learning about AI? 11% 20% 50% 69% Percentage of respondents whose understanding of AI has changed a lot or to a great extent in the past year 10% 20% 30% 40% 50% 60% 70% 80% 90% 0% Passives Experimenters Investigators Pioneers Percentage of respondents with an increased investment in AI in the past year 19% 62% 62% 88% Percentage of respondents with an increased investment in AI in the past year 

The State of AI 2019: Divergence Laggards are falling further behind in AI strategy 26 While more companies have an AI strategy, the proportion of laggards with an AI strategy is unchanged Source: S. Ransbotham, P. Gerbert, M. Reeves, D. Kiron, and M. Spira, “Artificial Intelligence in Business Gets Real,” MIT Sloan Management Review and The Boston Consulting Group, September 2018. • Laggards’ sense of urgency regarding AI is weakening. • The proportion of companies that have implemented an AI strategy has increased – but the proportion of laggards that have done so is unchanged. How are organisations planning for AI? Developing a strategy for AI is urgent for our organisation Percentage of all respondents We have a strategy for what we are going to do with AI on our organisation Percentage of respondents who expressed urgency for an AI strategy Somewhat or strongly agree Do not somewhat or strongly agree 2017 data 60% 39% 61% 67% 33% 69% 31% 85% 15% 40% 57% 14% 86% 50% 50% 63% 37% 90% 10% 43% Investigators Pioneers Experimenters Passives OVERALL 

The State of AI 2019: Divergence AI initiation is shifting from the C–suite to IT department 27 Initiation of AI projects has shifted from the C-Suite to the IT department Source: Gartner (June 2018) • Two years ago CEOs, CIOs or CTOs initiated two thirds of AI projects. • Today, just one in eight survey respondents highlight corporate leadership as the primary initiator of AI projects. Interest in AI has shifted from the C-suite to the IT department. 

The State of AI 2019: Divergence Companies prefer to buy, not build, AI 28 Nearly half of companies favour buying AI solutions from third parties Source: Gartner (June 2018). • Nearly half of companies favour buying AI solutions from third parties. • A third of companies intend to build custom solutions. Buy an AI solution and then customise it to our industry Buy an AI solution already customised to our industry Build custom solution internally using open source where possible Outsource custom solution development to third party Wait for AI to be embedded in our favoured software products Build custom solution internally Other 0 5 10 20 15 25 30 Sourcing strategies for AI initiaves Percentage of respondents 18 27 BUY = 44% BUILD = 33% 1 10 11 18 15 Communications = 32% Education = 25% Government = 25% 

Workers’ views vary widely regarding the impact of AI on their activities Source: Survey Analysis: How AI Will Impact Industries From the Workers’ Perspective, Gartner 2018 On balance, workers expect AI to decrease job security Source:Survey Analysis: How AI Will Impact Industries From the Workers’ Perspective, Gartner 2018 29 Workers are concerned about job security The State of AI 2019: Divergence • Workers expect AI to increase the safety, quality and speed of their work. • As companies’ AI agendas shift from revenue growth to cost reduction initiatives, workers are concerned about job security. 0 Fig. 19: Workers views regarding the impact of AI on their activities vary widely Percentage of respondents AI impact on jobs Variety % DECREASE % INCREASE Job security Customer contact Interesting Safety Co-worker collab Quality Pace of work 21 -20 14 -33 16 -22 15 -22 23 -18 38 -8 34 -10 34 -17 0 -20 20 40 Safety AI impact on jobs (net % of respondents) Quality Pace Interest Co-worker collaboration Customer contact Variety Job security How AI will impact industries from the workers' perspective 

The State of AI 2019: Divergence 2. The war for talent 

While supply of AI talent is increasing, demand for AI professionals continues to exceed supply. Winners and losers are emerging in the competition
 to capture talent. High job satisfaction is intensifying the war for talent. The war for talent The State of AI 2019: Divergence 31 

The State of AI 2019: Divergence Demand for AI talent has doubled in 24 months 32 • In the UK, job listings for AI roles have increased 485% since 2014 (Indeed). •Demand for AI talent has doubled in 24 months. •There is a gulf between demand and supply, with 2.3 roles available for every AI professional. AI-related job postings as a proportion of total job postings have doubled in 18 months Source: Indeed Fig. 1: AI-related job postings as a proportion of the total have doubled in 18 months (AI-related postings per million postings) 0 200 400 600 800 1000 1200 2015 2016 2017 2018 AI-related postings per million postings 

33 • Machine learning has become the top emerging field of employment in the United States. • Greater supply is being driven by: high pay; the inclusion of AI modules in university computer science courses; companies’ investment in staff training; and AI technology companies ‘pump priming’ the market with free resources. • Over time, AI tools offering greater abstraction will make AI accessible to less specialised developers. In the US, machine learning is the top emerging job Source: LinkedIn Technology companies are offering free educational AI resources Source: Google, Udacity Brand Partner Full Stack Developer Director of Data Science Unity Developer Big Data Developer Full Stack Engineer Customer Success Manager Sales Development Representative Data Scientist Machine Learning Engineer 0x 1x 2x 3x 4x 5x 6x 7x 8x 9x 10x Fig. 2: In the US, machine learning is the top emerging job Increase in individuals’ listed professions compared with five years ago Supply is increasing… 

• AI demands advanced competencies in mathematics, statistics and programming. • AI developers are seven times more likely to have a Doctoral degree than other developers. • Estimates vary; there may be as few as 22,000 highly-trained AI specialists (Element) and up to 300,000 AI researchers and practitioners within broader technical teams (Tencent). 60% of AI developers have a Master’s or Doctoral degree Source: Kaggle AI developers are seven times more likely to have a Doctoral degree than others Source: Kaggle, StackOverflow Fig. 5: 60% of AI developers have a Master’s or Doctoral degree (AI developers’ level of education) Master's degree Bachelor's degree Doctoral degree Other 42% 32% 16% 11% 0% 10% 20% 30% 40% 50% 60% Secondary school Some university Bachelor’s degree Master’s degree Doctoral degree AI Developers General Developers Developers (%) Fig. 6: AI professionals have a greater proportion of advanced degrees Developers’ levels of education 34 …but the pool of AI talent remains small The State of AI 2019: Divergence 

35 Over time, AI democratisation will mitigate talent shortages The State of AI 2019: Divergence • Over time, a larger talent pool and more accessible AI tools will alleviate much of the shortfall in AI talent. • Governments’ investment in education – in science, technology, engineering and mathematics (STEM) subjects – will be vital for countries to broaden their pools of AI talent. • AI courses and resources from universities and technology companies, and market demand, will also boost supply. • Development environments offering higher levels of abstraction will also reduce the burden on developers’ knowledge. In 2005, NumPy abstracted portions of required mathematics. In 2010 and 2015, Python libraries and TensorFlow progressively abstracted network development. 

• AI professionals are among the best paid developers and their salaries continue to increase. • Half enjoyed salary growth of 20% or more in the last three years. AI professionals are among the highest paid developers Source: StackOverflow Most AI professionals’ salaries have increased in the last three years Source: Kaggle Mobile developer Game or graphics developer Educator or academic reseracher Designer Front-end developer Database administrator QA/test developer System administrator Back-end developer Desktop or enterprise applications developer Full stack developer Embedded applications/devices developer Data or business analyst Data scientist or machine learning specialist DevOps specialist Engineering manager $0 $10 $20 $30 $40 Median annual salary ($000s) $50 $60 $70 $80 $90 Fig. 8: AI professionals are among the highest paid developers (Global average developer salaries by category) Global average developer salaries by category 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 0 Declined Unchanged (+/- 5%) Higher (+6% to +19%) Higher (+20% or more) Fig. 11: Most AI professionals’ salaries have increased in the last three years 3% 18% 19% 45% Compensation changes in the past three years for data scientists/machine learning engineers 36 Today, talent shortages are sustaining high salaries The State of AI 2019: Divergence 

• Data scientists enjoy among the greatest salary premium relative to their level of experience. AI professionals are paid highly relative to their level of experience Source: StackOverflow $40,000 $50,000 $60,000 $70,000 $80,000 $90,000 6 7 8 9 10 11 Average years of professional programming experience Median salary ($USD) Fig. 10: AI professionals are paid highly relative to their level of experience (Developer salary vs. years of professional programming experience) Developer salary vs. years of professional programming experience Number of respondents 10,000 20,000 DevOps specialist Engineering manager CTO/CEO/etc Product manager Data scientist Full-stack developer Embedded/devices developer Back-end developer QA or test developer Data or business analyst Mobile developer Game or graphics developer Educator or academic researcher Desktop of enterprise application developer System administrator Database administrator Designer Front-end developer Today, talent shortages are sustaining high salaries 37 The State of AI 2019: Divergence 

Technology and financial services firms are absorbing nearly 60% of AI talent Source: The Burtch Works Study – May 2018. N=2,212 . Shows distribution of data scientists by sector. 44% Technology 10% Other 14% Financial Services 6% Healthcare/ Pharma 2% Academia 1% Gaming 1% Government 5% Retail & CPG 9% Advertising/Marketing 8% Consulting Fig. 13: Technology and Financial Services absorb nearly 60% of AI talent (distribution of data scientists by industry) • The technology and financial services sectors are absorbing 60% of AI talent. • The ‘brain drain’ from academia to industry is real and will have mixed implications, catalysing AI’s impact while moving value from the public domain to select, private companies. Winners and losers are emerging in the war for talent 38 

• Three quarters of AI professionals are satisfied in their current role. • To optimise hiring and retention, companies should align roles to AI professionals’ primary motivators – learning opportunities, environment, access to preferred technologies and impact. Three quarters of AI developers are satisfied (6/10 or better) with their current role Source: Kaggle Learning, office environment and technologies are AI developers’ primary motivators Source: Kaggle Fig. 14: Three quarters of AI developers are satisfied (>6/10) with their current roles (AI developers’ job satisfaction) Respondents (%) AI developers’ job satisfaction (scale of 1 to 10) 0% 5% 10% 15% 20% 25% 1 2 3 4 5 6 7 8 9 10 0% 10% 20% 30% 40% 50% 60% 70% 80% Diversity Work remotely Leader reputation Company funding/financials Industry Publishing opportunties Commute Department Impact Job title Management Experience level Compensation Technologies Office environment Learning Fig. 15: Learning, office environment and the technologies they will use are AI developers’ primary motivators % respondents rating factor “very important” High job satisfaction is intensifying the war for talent 39 The State of AI 2019: Divergence 

• Company websites and technology job boards are less effective than engaging with recruiters, friends, family and colleagues, according to those already employed in the field. The most effective route into AI is engagement with recruiters Source: Kaggle 1. Recruiter 2. Friend, family, colleague 3. General job board 4. Other 5. Company website 6. Career fair or recruiting event 7. Tech job board EMPLOYED IN FIELD 1. Company website 2. Tech job board 3. General job board 4. Friend, family, colleague 5. Recruiter 6. Career fair or recruiting event 7. Other ENTERING FIELD Fig. 16: The most effective route into AI work is engagement with recruiters How do you look for, or find, work? 40 The State of AI 2019: Divergence New talent follows sub-optimal paths to employment 

The State of AI 2019: Divergence 3. The advance of technology 

The advance of technology The State of AI 2019: Divergence 42 Advances in AI technology are creating new possibilities. Custom silicon is enabling a new generation of AI hardware. Emerging software techniques – reinforcement learning, transfer learning and generative AI – are delivering breakthroughs in multiple domains and decoupling progress from the constraints of human experience. 

• Vendors are optimising or customising hardware to support the use of popular deep learning frameworks, enabling faster system training and performance with common AI frameworks. • NVIDIA’ Tesla GPUs accelerate matrix computations to optimise deep learning on a range of frameworks. Certain neural networks can be trained in one third of the previous time and operate 4x faster. Tesla GPUs enable suitable neural networks to be trained in a third of the previous time Source: NVIDIA Tesla GPUs allow suitable neural networks to operate four times faster than previously Source: NVIDIA 43 Tensor architectures are accelerating deep learning The State of AI 2019: Divergence 8X Tesla P100 8X Tesla V100 0 4 8 12 16 Time to solution in hours (Lower is better) Server config: Dual Xeon E5-2699 v4 GHz | 8X NVIDIA® Tesla® P100 or V100 | ResNet-50 Training on MXNet for 90 Epochs with 1.28M ImageNet Dataset Title TBC 15.5 hours 5.1 hours Tesla P4 1X CPU Tesla V100 0 10X 20X 27X 7X 30X 40X 50X Performance normalised to CPU Workload: ResNet-50 | CPU: 1X Xeon E5-2690v4 @ 2.6 GHz | GPU: Add 1X Tesla P4 or V100 Title TBC 

Tensor architectures are accelerating deep learning 44 • Google’s Tensor Processing Unit (TPU) is an application- specific integrated circuit (ASIC) designed specifically to accelerate AI workloads on the popular TensorFlow framework. • Pods of next-generation TPUs can reduce training time for select algorithms from
 five hours to seven minutes. Google’s next-generation TPUs reduce further the time required to train an algorithm Source: Google Title TBC ResNet-50 302.0 minutes 183.0 minutes 60.0 minutes 7.1 minutes TPU v2 TPU v3 TPU v3 Pod TPU v3 Pod NMT SSD 40.4 minutes 26.3 minutes 9.7 minutes 96.9 minutes 74.9 minutes 17.8 minutes Number of chips 4 4 16 4 256 4 

• Custom silicon, designed from inception for AI, offers transformational performance, capability similar to existing systems for a fraction of the power or space, and greater versatility. • Graphcore’s intelligence processing unit (IPU) combines a bespoke, parallel architecture with custom software to offer greater performance than existing systems. Graphcore’s IPU is designed, from inception, for AI Source: Graphcore Graphcore’s IPU could deliver 200-fold performance improvements in selected tasks Source: Graphcore 45 The post-GPU era: custom silicon offers new possibilities The State of AI 2019: Divergence GPU (<2ms latency) GPU (<7ms latency) IPU (<7ms latency) GPU (<5ms latency) IPU (<5ms latency) IPU (<2ms latency) 0 10000 20000 242X 182X 30000 40000 50000 60000 70000 Inferences per second Single layer inference 

• A ‘barbell’ effect is emerging as a new class of AI hardware is optimised for edge computing instead of the data centre. • Edge computing moves the processing of data from the cloud to the ‘edge’ of the internet – to devices where data is created: autonomous vehicles; drones; sensors and IoT devices. • Increasingly, edge computing is required as devices proliferate, and issues of connectivity and latency demand on-device processing in multiple contexts. 46 Custom silicon is taking AI to the edge The State of AI 2019: Divergence In 2019, as well as enabling next generation AI in the cloud, custom silicon will transform AI at the edge by coupling high performance with low power consumption and small size. 

• As quantum computing matures it will create profound opportunities for progress in AI. • Quantum-powered AI will enable humanity to address previously intractable problems, from personalised medicine to climate change. • While nascent, quantum computing is advancing rapidly. Google is developing quantum neural networks. In 2018, an Italian team of researchers developed a functioning quantum neural network on an IBM quantum computer. 47 Quantum computing is unlocking profound opportunities The State of AI 2019: Divergence Interested in learning more about quantum computing? Listen to ‘Quantum Leap’ – Episode 6 – of our ‘Beyond the Hype: AI’ podcast:
 => bit.ly/2LrNhn1 (Web) => apple.co/2GwMXEU (iTunes) 

Reinforcement learning is enabling powerful AI agents 48 Reinforcement learning enabled AlphaGo Zero, a system developed by DeepMind to play Go, to achieve unrivalled capability after 40 days of play Source: Google DeepMind • Instead of learning from training data, reinforcement learning (RL) systems learn through exploration. Progress towards a specified goal is rewarded and reinforced. • Developments in RL are enabling the creation of powerful AI agents. • AlphaGo Zero (Oct 2017), an RL system developed to play Go, achieved unparalleled capability within 40 days by playing against itself. AlphaZero (Dec 2017) can play a range of games at even higher levels. Elo Rating Days -2000 -1000 0 1000 2000 3000 4000 5000 Title TBC 0 5 10 15 20 25 30 35 40 40 days AlphaGo Zero supasses all other versions of AlphaGo and, arguably, becomes the best Go player in the world. It does this entirely from self-play, with no human intervention and using no historical data. AlphaGo Zero 40 blocks AlphaGo Lee AlphaGo Master 

Reinforcement learning is enabling effective multi-agent collaboration Source: OpenAI. AI agents playing Defence of the Ancients 2 (Dota2) Reinforcement learning enabled the OpenAI 5 team to surpass rapidly most human teams Source: OpemAI 49 Reinforcement learning is enabling multi-agent collaboration The State of AI 2019: Divergence Drafting Picking Composition Mirror Death Prophet, Witch Doctor, Gyrocopter, Earthshaker, Fidehunter Mirror Necrophos, Lich, Crystal Maiden, Viper, Sniper 2,000 3,000 4,000 5,000 6,000 22 April 6 May OpenAI Five OpenAI Dev Team Blitz + Audience Amateur Team Semi-Pro Team Test Team A Test Team B Caster Team 20 May 3 June 17 June 1 July 15 July 29 July 12 August Reinforcement learning enabled the OpenAI 5 team to rapidly surpass the performance of most human teams Approximate Team MMR Date (2018) • Developments in RL are enabling groups of agents to interact and collaborate with each other. • OpenAI 5, an RL-powered team of agents developed to play the game Defence of the Ancients 2 (DOTA2), improved rapidly and can defeat all but the top professional human players. 

Reinforcement learning decouples progress from human knowledge 50 • RL is well suited to creating agents that can perform autonomously in environments for which we lack training data. • Progress in RL is significant because it decouples system improvement from the constraints of human knowledge. • RL enables researchers to “achieve superhuman performance in the most challenging domains with no human input” (DeepMind). The State of AI 2019: Divergence 

Transfer learning can offer strong initial performance, faster improvement and better 
 long-term results Source: Torry, Shalvik Interest in transfer learning has grown 7-fold in 24 months Source: Google trends 51 Transfer learning enables more adaptable systems • Transfer learning (TL) enables programmers to apply elements learned from previous challenges to related problems. • Interest in TL has grown seven-fold in 24 months and is enabling a new generation of systems with greater adaptability. • TL- powered models are improving the state of the art in language processing in areas of universal utility – text classification, summation, text generation, question answering and sentiment analysis. Interest in transfer learning has grown 7-fold in 24 months 1.0 2.0 3.0 4.0 5.0 6.0 7.0 8.0 APR-13 JUN-13 AUG-13 OCT-13 DEC-13 FEB-14 APR-17 JUN-17 AUG-17 OCT-17 DEC-17 FEB-18 APR-18 JUN-18 AUG-18 OCT-18 DEC-18 FEB-19 APR-14 JUN-14 AUG-14 OCT-14 DEC-14 FEB-15 APR-15 JUN-15 AUG-15 OCT-15 DEC-15 FEB-16 APR-16 JUN-16 AUG-16 OCT-16 DEC-16 FEB-17 Relative interest in transfer learning (April 2013 = 1.0) Performance Transfer learning can offer strongerinitial performance, faster improvement and better long-term results Training with transfer without transfer higher start higher slope higher asymptote 

52 Transfer learning is delivering leaps in language AI The State of AI 2019: Divergence • ELMo (‘Embeddings from Language Models’, March 2018): improved the state of the art for a broad range of natural language tasks including Q&A and sentiment analysis. • ULMFiT (‘Universal Language Model Fine-tuning for Text classification’, May 2018): underscored that TL can be applied to language processing tasks and introduced techniques for fine-tuning language models. With 100 labelled examples, ULMFiT matched the performance of systems trained with 100-fold more data. • OpenAI (mid-2018): Impressive results on a diverse range of language tasks from a single starting point; task-agnostic model outperformed models with architectures crafted for specific tasks including question answering. • BERT (Bidirectional Encoder Representations from Transformers, Oct 2018): Open-sourced by Google, an RL-based language processor achieved state-of-the-art results on 11 natural language benchmarks. 

53 Transfer learning will unlock further advances in AI The State of AI 2019: Divergence • Transfer learning is enabling complex systems to interact with the real world.
 Training AI systems, such as autonomous vehicles, can be laborious, expensive or dangerous. Simulation enables emulation of environments instead of capturing real-life data. Applying transfer learning enables learnings from a simulation to be applied to real-world assets. • Transfer learning offers adaptability and progress towards AGI.
 By offering greater adaptability, TL supports progress towards artificial general intelligence (AGI)
 i.e. systems that can undertake any intellectual task a human can perform.
 While AGI is far from possible with current AI technology, developments in TL are enabling progress. 

“I think transfer learning is the key to general intelligence. And I think the key to doing transfer learning will be the acquisition of conceptual knowledge – knowledge that is abstracted away from perceptual details of where you learned it, so you can apply it to a new domain.” Demis Hassabis, DeepMind The State of AI 2019: Divergence 54 Transfer learning will unlock further advances in AI 

GANs’ ability to create lifelike media has rapidly improved Source: Goodfellow et al, Radford et al, Liu and Tuzel, Karras et al, https://bit.ly/2GxTRot GANs can generate artificial images that appear real (none of these individuals exist) Source: NVIDIA 55 Generative AI will transform media and society The State of AI 2019: Divergence • Generative Adversarial Networks (GANs) enable the creation of artificial media, including pictures and video, with exceptional fidelity. • GANs will deliver transformational benefits in sectors including media and entertainment, democratising the creation of high quality content at scale and low cost. GANs will also present profound challenges to societies – beware ‘fake news 2.0’. 

Generative AI: two networks working in opposition 56 GANs operate with two networks working in opposition Source: Naoki Shibuya • GANs operate by two networks – a ‘generator’ and ‘discriminator’ – working in opposition to create increasingly lifelike media. • The generator receives a random input, such as a matrix of numbers, and follows a series of mathematical transformations to convert the input into a picture. • The discriminator assesses whether the image received from the generator is authentic or has been artificially generated, with a feedback cycle between the networks promoting progress. DISCRIMINATOR GENERATOR Training data Latent sample -0.19972104, -0.27617624, -0.7449326, -0.02522890, -0.86428853, -0.84710504, -0.41719925, -0.88549566, -0.42638235, -0.0455994, -0.305852, -0.60752668, -0.14700781, -0.26471074, -0.71651508, -0.65559716, -0.71335986, -0.82961057, -0.81311934, -0.42092858, -0.42457545, -0.39863341, -0.26192929, -0.18518651, Generated image 1 (Real) 0 (Fake) Title TBC The State of AI 2019: Divergence 

57 Generative AI has use beyond pictures and video The State of AI 2019: Divergence • While GANs are frequently used to create images, their utility is broader. GANs enable: - Alternative media: create music or text in the style of particular individuals. - System Training: improve the training of AI classification systems. - Data manipulation: Add/remove features in data by combining GANs with autoencoding techniques. - Data normalisation: Normalise data from different sources. - Network security: Detect anomalies in network access or activity. - Data creation: Produce additional training data to improve the accuracy of AI classifiers. 

The State of AI 2019: Divergence 4. The Disruptors:
 Europe’s AI startups 

1 in 12 European startups is now an ‘AI startup’
 putting AI at the heart of its value proposition. Europe’s 1,600 AI startups are maturing, bringing creative destruction to new industries and navigating unique capital dynamics. While the UK is the powerhouse of European AI, Germany and France may extend their influence. Entrepreneurs’ greatest challenges are access to data, recruiting talent and moving AI from ‘lab to live’. The Disruptors: Europe’s AI startups The State of AI 2019: Divergence 59 

Europe is home to 1,600 early stage AI software companies 60 • We individually reviewed the activities, focus and funding of 2,830 companies classified 
 by popular tools as AI startups, in the 13 EU countries most active in AI. • Together, these countries – Austria, Denmark, Finland, France, Germany, Ireland, Italy, the 
 Netherlands, Norway, Portugal, Spain, Sweden and the UK – also comprise 90% of EU GDP. • In approximately 60% of cases – 1,580 companies – there was evidence of AI material to
 a company’s value proposition. The State of AI 2019: Divergence 

One in twelve new European startups is an AI company Source: MMC Ventures (2018 data to October) • In 2013, one in fifty new startups embraced AI. • Today, one in twelve put AI at the heart of their value proposition. • We have entered the era of AI entrepreneurship. AI entrepreneurship is becoming mainstream 61 The State of AI 2019: Divergence 0 1% 2% 3% 4% 5% 6% 7% 8% 9% 0.0pp 0.2pp 0.4pp 0.6pp 0.8pp 1.0pp 1.2pp 1.4pp 1.6pp 1.8pp 2011 2010 2012 2013 2014 2015 2016 2017 2018 AI startups as % of all startups founded each year AI startups as % of all startups founded each year AI startups as % of all startups founded each year (left axis) YoY change in percentage points (right axis) YoY change in percentage points 

• Two years ago, just one in 20 AI startups were ‘growth’-stage companies that had passed beyond Angel and Seed funding rounds. • Today, one in six European AI companies is a ‘growth’-stage company with over $8m of funding. • Expect: acquisitions to recycle capital and talent; startups competing with ‘scale-ups’ as well as incumbents; and increasing competition for talent. Six in ten startups are at the Angel or Seed stages Larger ecosystems are typically more mature; Spain is an exception Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn The European AI ecosystem is maturing 62 Six in ten AI startups are at the Angel or Seed stages Angel 27% Early Stage 28% Seed 29% Growth 16% 0% 20% 40% 60% 80% 100% Fig. X: Larger ecosystems are typically more mature; Spain is an exception 31% 26% 28% 29% 28% 28% 34% 25% 19% 19% 27% 27% 27% 16% 16% 39% 21% 35% 32% 32% 11% 38% 24% 9% 6% 18% 26% All Seed Early Stage Growth Angel UK France Germany Spain Sweden Finland 25% Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn.
 Funding categories: Angel (<$0.5m), Seed ($0.5-2.0m), Early Stage ($2.0-8.0m), Growth (>$8.0m) 

With twice as many AI startups as any other country, the UK is the powerhouse of European AI entrepreneurship Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn • The UK is the heartland of European AI with a third of the continent’s AI startups. • UK strengths include: talent from a quarter of the world’s top 25 universities; extensive venture capital; many successful AI scale-ups and exits; and a global financial services hub. • Germany and France are thriving hubs where high quality talent, increasing volumes of capital and an expanding roster of successful AI companies are creating feedback loops of growth and investment. The UK is the powerhouse of European AI 63 479 UK 196 Germany 217 France 166 Spain 75 Ireland 66 Italy 45 Portugal 43 Austria 32 Norway 36 Denmark 49 Finland 73 Sweden 103 Netherlands Fig. X: The UK is the heart of European AI entrepreneurship 

The ‘UK 500’: A map of early stage UK AI companies 64 The State of AI 2019: Divergence Cambridge Quantum Computing UK AI Landscape (Early stage companies) ANGEL / SEED EARLY STAGE / GROWTH Skim Technologies Instadeep Memgraph Seldon Invacio Grakn Labs Hadean Graphcore Fetch.AI Academy of Robotics Evolve Dynamics ANGEL / SEED FiveAI EARLY STAGE / GROWTH Oxbotica Latent Logic Focal Point Positioning Sky-Futures Accelerated Dynamics Baro Vehicles Headlight AI Humanising Autonomy Intelligent Robots Machines With Vision Predina Wayve React AI Third Space Auto ANGEL / SEED EARLY STAGE / GROWTH SLAMcore MicroBlink Spectral Edge WeSee Blue Vision Labs Emteq Alchera Technologies FaceSoft TouchByte Vize.ai Visio Ingenii Machine Medicine Synthesia EARLY STAGE / GROWTH Telectic EARLY STAGE / GROWTH Audio analytic Ditto AI Mind Foundry Prodo.ai ANGEL / SEED Bottr.co Chatbot Agency Prowler.io Neuron soundware CloudNC Senseye QiO Technologies EARLY STAGE / GROWTH ANGEL / SEED EARLY STAGE / GROWTH Ai Build Flexciton Materialize.X Thingtrax ANGEL / SEED Mouldbox Houseprice.AI PerchPeek Pupil ANGEL / SEED EARLY STAGE / GROWTH AskPorter Skyscape Ecosync Brytlyt Hivemind Instrumental Jukedeck Myrtle Software Mogees Popsa Spirit AI ANGEL / SEED Capito Systems Kami.ai Poly AI CrowdEmotion DAM Good Media AI Music Antix Let's Enhance Lobster Logically infloAi Kompas DeepAR Deep Innovations Gameway Factmata Pimloc Thingthing Snaptivity B2B Angel/Seed: <$2m Early stage/Growth: >$2m Funding category New company (since 2017) B2C KEY iProov Speechmatics RETAIL OTHER LANGUAGE & SOUND COMPUTER VISION AI INFRASTRUCTURE Core Technologies AUTONOMOUS SYSTEMS Sector Cortexica Dressipi Artlimes Cocoon Emotech Mallzee Metail Snap Tech Spoon Guru Thread Aura Vision Boldmind ANGEL / SEED EARLY STAGE / GROWTH Hoxton Analytics JCC Bowers Measmerize Cadouu CartMe Orpiva Pasabi Proximus See Fashion ThirdEye Save your wardrobe Olvin EDITED PROPERTY & REAL ESTATE Gousto Winnow Nuritas Henchman ANGEL / SEED EARLY STAGE / GROWTH Dinabite Limited IntelligentX VideraBio FOOD & BEVERAGE MANUFACTURING MEDIA & ENTERTAINMENT Function Energi Mine SenSat Limejump nPlan EARLY STAGE / GROWTH ClauseMatch Eigen Technologies Luminance Cognitiv+ ThoughtRiver DoNotPay ANGEL / SEED EARLY STAGE / GROWTH Crowd Connected Verv by Green Running Calipsa Disperse.io ANGEL / SEED Green Running Grid Edge NumberEight TravelAI OpenCapacity Wluper Rovco Ginie.ai INFRASTRUCTURE & TRANSPORT Cambridge Medical Robotics Exscientia GTN Drayson Technologies BrainWaveBank BenevolentAI Babylon Health Amiko BioBeats Cydar Medical Deontics eCUORE Antiverse Biorelate Auro ANGEL / SEED EARLY STAGE / GROWTH LabGenius Lifebit Cambridge Humanae Cambridge Bio-Augmentation Systems Aigenpulse Causaly Chronomics HealthUnlocked Healx Kheiron Medopad Snap40 Synthace Thriva Your.MD Visulytix Viz Sentireal ThinkSono Sime Diagnostics Transformative HEALTH & WELLBEING LAW FitWell InnersightLabs InsideDNA Kraydel Kaido Kiroku OME Health Resurgo Genetics PetaGene Optellum Peptone ANGEL / SEED EARLY STAGE / GROWTH Quantemplate Quantexa Vortexa ArrayStream AdviceGames Artificial Labs Banked Blue Lion Research Bud Abaka FINANCIAL SERVICES Oxademy Synap BridgeU Gamar Kwiziq ANGEL / SEED EARLY STAGE / GROWTH Century Lingumi Lingvist EARLY STAGE / GROWTH Global Surface Intelligence Hummingbird Technologies Dogtooth Technologies HeraSpace Observe ANGEL / SEED Cervest KisanHub Xihelm Optimal Labs AGRICULTURE EDUCATION Tractable Procensus PriceHubble Plum Kirontech Essentia Analytics Arkera Floodlash BMLL Technologies Digital Contact FriendlyScore ForwardLane Cleo AI Cytora Mosaic Smart Data Multiply TradeTeq Acuity Trading AlgoDynamix Brolly Alpha-i CUBE DealX Ducit.ai Knox EA G-Research Metafused Digital Clipboard Financial Network Analytics Chip EnAlgo Oseven Telematics Proportunity Yedup Zoral Sns Analytics ProvidensAI Torafugu Tech Win FX Financial Innovation TradeRiser Spixii RightIndem UK AI Landscape (Early stage companies) ANGEL / SEED EARLY STAGE / GROWTH Skim Technologies Instadeep Memgraph Seldon Invacio Grakn Labs Hadean Graphcore Fetch.AI Academy of Robotics Evolve Dynamics ANGEL / SEED FiveAI EARLY STAGE / GROWTH Oxbotica Latent Logic Focal Point Positioning Sky-Futures Accelerated Dynamics Baro Vehicles Headlight AI Humanising Autonomy Intelligent Robots Machines With Vision Predina Wayve React AI Third Space Auto ANGEL / SEED EARLY STAGE / GROWTH SLAMcore MicroBlink Spectral Edge WeSee Blue Vision Labs Emteq Alchera Technologies FaceSoft TouchByte Vize.ai Visio Ingenii Machine Medicine Synthesia EARLY STAGE / GROWTH Telectic Audio analytic Prodo.ai Prowler.io Brytlyt Hivemind ANGEL / SEED Capito Systems Poly AI B2B Angel/Seed: <$2m Early stage/Growth: >$2m Funding category New company (since 2017) B2C KEY Spe LANGUAGE COMPUTER VISION AI INFRASTRUCTURE Core Technologies AUTONOMOUS SYSTEMS Sources: MMC Ventures, Beauhurst, Crunchbase, Tracxn Additions or corrections? Email us at [email protected] 

The ‘UK 500’: A map of early stage UK AI companies 65 The State of AI 2019: Divergence ANGEL / SEED Ai Build Flexciton Materialize.X Thingtrax ANGEL / SEED Mouldbox Houseprice.AI PerchPeek ANGEL / SEED AskPorter Skyscape Ecosync CrowdEmotion DAM Good Media AI Music Antix Let's Enhance Lobster Logically infloAi Kompas DeepAR Deep Innovations Gameway Factmata Pimloc Thingthing Snaptivity Cybertonica Fraugster Ravelin Featurespace ANGEL / SEED EARLY STAGE / GROWTH Fractal Labs Aire Rimilia Pace Fluidly ANGEL / SEED EARLY STAGE / GROWTH FRAUD DETECTION FINANCE HUMAN RESOURCES PredictiveHire hackajob GoSay Grad DNA Filtered Headstart App Beamery Qlearsite Rotageek Saberr StatusToday ANGEL / SEED EARLY STAGE / GROWTH Potentially Metaview MeVitae Stitched TalentHunter.AI Wevolve JamieAi InteriMarket Human Darktrace RepKnight Senseon Alchemy Data Cyberlytic Barac Encode Corax CyberSparta Elemendar ANGEL / SEED EARLY STAGE / GROWTH Honeycomb Technologies VChain Technology Cybershield CYBERSECURITY Aistemos Amplyfi EARLY STAGE / GROWTH Black Swan Data Cazana Bird.i Gyana Logical Glue Callsign Behavox Audit XPRT Berry Technologies CoVi Analytics Exonar Eyn AimBrain Hazy Onfido Tessian Traydstream Limited Sum&Substance ANGEL / SEED EARLY STAGE / GROWTH WaymarkTech BotsAndUs DigitalGenius Gluru Hutoma ANGEL / SEED EARLY STAGE / GROWTH Action.AI Enterprise Bot True AI Synthetix Constellation AI Humley Sentient Machines Samim.ai Massive Analytic Peak Quorso Rezatec Ripjar Semantic Evolution Signal Media Simudyne ANGEL / SEED 10x Airfinity causaLens Analytics Intelligence Chorus Intelligence Data quarks Flumes Hertzian Policy Radar Reportbrain SeeQuestor Sensing Feeling Satalia Singular Intelligence Terrabotics illumr Krzana Kite Edge illumr Ohalo Migacore Technologies Oxford Semantic Technologies Hello Soda CUSTOMER SERVICE BI & ANALYTICS COMPLIANCE Artlimes Aura Vision Boldmind ANGEL / SEED Hoxton Analytics JCC Bowers Measmerize Cadouu CartMe Orpiva Pasabi Proximus See Fashion ThirdEye Save your wardrobe Olvin Decibel Insight Fresh Relevance Brandwatch LoopMe MiQ Buzz Radar Admedo Fospha ADTYPE adverttu Artios ANGEL / SEED EARLY STAGE / GROWTH Carsift Chattermill Concured Creative AI Crystal Apps CustomSell Digital MR DaVinci11 MediaGamma Perfect Channel Storystream Realeyes Qubit Jampp FindTheRipple Firedrop Mercanto Metageni Mobile Acuity Nudgr Phrasee Ignition Ai Maybe* Advizzo ArtuData Aiden Bibblio rais Platform360 Vaix Swogo MARKETING & ADVERTISING ANALYTICS/OPTIMIZATION TARGETING BoomApp ANGEL / SEED EARLY STAGE / GROWTH Selerio Echobox Adoreboard Colourtext ANGEL / SEED EARLY STAGE / GROWTH Genus SentiSum Big Data for Humans Codec ANGEL / SEED EARLY STAGE / GROWTH AshTV Popcorn Metrics Otus Labs Personalyze Idio iotec Pixoneye Permutive DataSine Visii ANGEL / SEED EARLY STAGE / GROWTH Donaco SENTIMENT ANALYSIS AUGMENTED CONTENT PURCHASE DISCOVERY/ RECOMMENDATION Function IT Cardinality 4th Office Mettrr Rainbird Technologies Global App Testing re:infer Recordsure Redsift Automorph Ampliphae ANGEL / SEED EARLY STAGE / GROWTH Aria Networks BigHand Autto Beneficiary.io CompareSoft Automated Intelligence Diffblue Trint Vivid-q Context Scout Retechnica Nexus Frontier Tech Cyanapse Digital Taxonomy jClarity Fantoo Fedr8 Fraim Linguamatics Mudano Rossum Skipjaq Thingful Unity {Cloud}Ware TextRazor Synthesized Spot Intelligence Spotlight Data PROCUREMENT Previse keelvar Matchdeck ANGEL / SEED EARLY STAGE / GROWTH SALES R&D FeedStock Klydo Sparrho Patsnap GlassAI ANGEL / SEED EARLY STAGE / GROWTH DueDil Cognism GrowthIntel Artesian Solutions Conversity Kluster Intelligence SalesSift Synoptic Technologies ANGEL / SEED EARLY STAGE / GROWTH Netz EnigmaMS UK AI Landscape (Early stage companies) ANGEL / SEED EARLY STAGE / GROWTH Skim Technologies Instadeep Memgraph Seldon Invacio Grakn Labs Hadean Graphcore Fetch.AI Academy of Robotics Evolve Dynamics ANGEL / SEED FiveAI EARLY STAGE / GROWTH Oxbotica Latent Logic Focal Point Positioning Sky-Futures Accelerated Dynamics Baro Vehicles Headlight AI Humanising Autonomy Intelligent Robots Machines With Vision Predina Wayve React AI Third Space Auto ANGEL / SEED EARLY STAGE / GROWTH SLAMcore MicroBlink Spectral Edge WeSee Blue Vision Labs Emteq Alchera Technologies FaceSoft TouchByte Vize.ai Visio Ingenii Machine Medicine Synthesia EA A P Prowler.io Brytlyt Hivemind A C P B2B Angel/Seed: <$2m Early stage/Growth: >$2m Funding category New company (since 2017) B2C KEY L COMPUTER VISION AI INFRASTRUCTURE Core Technologies AUTONOMOUS SYSTEMS Sector EARLY STAGE / GROWTH FINANCIAL SERVICES EARLY STAGE / GROWTH EARLY STAGE / GROWTH AGRICULTURE EDUCATION Sources: MMC Ventures, Beauhurst, Crunchbase, Tracxn Additions or corrections? Email us at [email protected] 

France and Germany are increasing their share of European AI entrepreneurship Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn The dynamics European AI entrepreneurship are in flux 66 The State of AI 2019: Divergence UK France Germany Spain Ireland Netherlands Sweden Italy Finland Portugal Austria Denmark Norway 0% 5% -5% -10% 10% 15% 20% 25% 30% 35% 40% Fig. X: France and Germany are increasing their share of European AI entrepreneurship Share of European AI startups Change in share over the last six years (% points) • While the UK remains the powerhouse of European AI, its share of Europe’s AI startups (by volume) has slightly decreased in recent years. • Brexit could accelerate this dynamic. If free movement of workers between the EU and UK ends, visas are unforthcoming or rhetoric is unwelcoming, the UK’s access to talent could be inhibited. • France, Germany and other hubs may extend their influence in the decade ahead, spreading the benefits of entrepreneurship more evenly
 across Europe. 

Italy, Sweden and Germany ‘punch above their weight’
 in core technology • Two thirds of of Europe’s ‘core technology’
 (sector-agnostic) AI startups are located in the
 UK, Germany and France. • However, adjusting for countries’ ‘size’ – their number of AI startups – Italy, Sweden and Germany ‘punch above their weight’ and
 are core technology hubs. The European core AI technology hubs 67 The State of AI 2019: Divergence Italy Sweden Germany Finland Austria Denmark Norway Spain UK France Netherlands Ireland Portugal 0 5% 10% 15% 20% 25% 30% Fig. X: Core technology startups as % of AI startups Core technology startups as % of AI startups Average across all countries % of Europe’s Core Tech 25% 30% Fig. X: Core technology startups as % of AI startups Core technology startups as % of AI startups Average across all countries % of Europe’s Core Tech Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn 

• Nine in ten of Europe’s 1,600 AI startups sell to other businesses (B2B). • Just one in ten AI startups sells directly to consumers (B2C). • But B2C AI is on the rise. In 2018, a quarter of new AI startups were B2C. • B2C entrepreneurs are mitigating or circumventing the ‘cold start’ data challenge. Nine in ten AI startups are B2B Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn B2C AI is on the rise – a quarter of new AI companies are B2C Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn Nine in ten AI startups are B2B, but B2C AI is on the rise 68 The State of AI 2019: Divergence Fig. X: Nine in ten of the AI startups in our universe are B2B B2B B2C 87% 13% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2011 2010 pre 2010 2012 2013 2014 2015 2016 2017 2018* Fig. X: The share of new AI startups focusing on B2C is increasing B2C B2B 90% 90% 94% 90% 87% 87% 90% 87% 83% 75% 10% 10% 6% 10% 13% 13% 10% 13% 17% 25% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2011 2010 pre 2010 2012 2013 2014 2015 2016 Fig. X: The share of new AI startups focusing on B2C is increasing B2C B2B 90% 90% 94% 90% 87% 87% 90% 87% 10% 10% 6% 10% 13% 13% 10% 13% 

• There is a higher proportion of B2C AI companies in sectors where data is readily available. • The sectors attracting the highest proportion of new AI startups have been: finance;
 health & wellbeing; and media & entertainment. Media, finance, and health attract a higher percentage of B2C companies Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn Half of new AI startups target the finance, health or media sectors Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn , Startups founded in the last 24 months. B2C AI is growing in sectors with available data 69 The State of AI 2019: Divergence Finance Infrastructure & Utilities Health & Wellbeing Media & Entertainment Retail Transport & Travel 0% 20% 40% 60% 80% 100% Fig. X: B2B/C dynamics by sector 79% 69% 53% 86% 73% 74% 21% 31% 47% 14% 27% 26% B2C B2B 5% 10% 15% 20% 25% 0% Finance Health & Wellbeing Media & Entertainment Retail Transport & Travel Fig. 5: Almost one in four of the sector AI startups founded in 2017/8 targets the financial services space 23% 17% 10% 10% 9% 

• Nine in ten AI startups address a need in a specific ‘vertical’ (business function or sector). • Just one in ten is developing a core, sector-agnostic ‘horizontal’ AI technology. • The proportion of core technology providers will remain modest as global technology companies offer an extensive and expanding suite of core AI technologies. One in ten AI startups focuses on core tech Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn The proportion of ‘horizontal’ core technology providers has remained consistent over time Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn , Startups founded in the last 24 months. AI entrepreneurship remains vertically-focused 70 Fig. X: One in ten AI startups focuses on core tech Core technologies Vertical (Sector/Function) 88% 12% 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2013 2014 2015 2016 2017 Fig. X: Vertical/Horizontal dynamics of new AI startups 43% 41% 44% 44% 37% 11% 15% 11% 10% 11% 45% 44% 45% 46% 52% Sectors Functions Core Technologies 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2013 2014 2015 2016 2017 43% 41% 44% 44% 37% 11% 15% 11% 10% 11% 45% 44% 45% 46% 52% Sectors Functions Core Technologies 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2013 2014 2015 2016 2017 Fig. X: Vertical/Horizontal dynamics of new AI startups 43% 41% 44% 44% 37% 11% 15% 11% 10% 11% 45% 44% 45% 46% 52% Sectors Functions Core Technologies 0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% 2013 2014 2015 2016 2017 Fig. X: Vertical/Horizontal dynamics of new AI startups 43% 41% 44% 44% 37% 11% 15% 11% 10% 11% 45% 44% 45% 46% 52% Sectors Functions Core Technologies 

• More AI startups – one in five – serve the health & wellbeing sector than any other. Healthcare is a focal point for AI entrepreneurship. • In the coming decade, developers will have a greater impact on the future of healthcare than doctors. • Activity is thriving given profound new opportunities for process automation and cost reduction and a tipping point in stakeholders’ openness to innovation. More AI startups – one in five – serve the health & wellbeing sector than any other Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn Health & wellbeing is a focal point for AI entrepreneurship 71 21% Health & Wellbeing 18% Finance 12% Media & Entertainment 6% Education 5% Agriculture 11% Retail 8% Transport & Travel 8% Infrastructure & Utilities Fig. X: One in five AI sector startups in our universe focuses on Health & Wellbeing 

The UK is the heartland of European healthcare AI Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn • The UK is home to a third of Europe’s health & wellbeing AI startups. • UK entrepreneurs benefit from: leading universities for medicine and high quality teaching hospitals; successful healthcare scale-ups stimulating talent; and increasing expenditure and openness to innovation within the National Health Service (NHS). The UK is the heartland of European healthcare AI 72 Fig. X: Share of European health & well being startups 35% 1% 

More startups – one in four – serve the marketing function than any other Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn • Among AI startups serving a business function, more – a quarter – serve the marketing function
 than any other. • Customer service and IT departments also enjoy a rich ecosystem of suppliers. Marketing and customer service teams are well served 73 The State of AI 2019: Divergence 23% Marketing 16% IT 16% Customer Service 9% Operations 7% Sales 4% Compliance 8% BI & Analytics 8% HR Fig. X: Marketing accounts for one in four AI function startups 

The UK has four in ten of Europe’s AI marketing startups (% of European AI marketing startups) Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn France is Europe’s hub for AI Customer Service (% of European AI Customer service startups) Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn The UK leads in marketing AI, France for customer service 74 The State of AI 2019: Divergence Fig. X: Share of European customer service startups 22% 1% Fig. X: Share of European marketing startups 39% 1% • The UK contributes half of Europe’s AI marketing startups. • France is Europe’s hub for customer service AI with a fifth of Europe’s customer service AI startups. 

• While currently underserved, the operations 
 function is benefitting from an influx of new,
 AI-led startups in the last 24 months. • AI is expanding the ‘envelope’ of automation – 
 the breadth and value of processes susceptible
 to digital mechanisation and robotic process 
 automation (RPA). • Over time, domain-focused vendors may build 
 defensibility through domain expertise, workflow 
 integrations, data network effects and referenceability. An influx of AI startups is driving process automation 75 The State of AI 2019: Divergence Among new startups addressing a business function, one in seven serve operations teams Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn 5% 10% 15% 20% 25% 0% Customer service Marketing Operations IT HR Fig. X: Almost one in five of the function AI startups founded in 2017/8 focuses on customer service % AI start-ups founded since 2017 serving a business function 20% 15% 15% 13% 13% 

Since 2015, AI companies have raised larger investment rounds Source: Crunchbase, MMC Ventures AI companies are raising larger rounds at all stages of maturity Source: Crunchbase, MMC Ventures AI companies raise larger amounts of capital 76 • AI companies raise larger volumes of capital than traditional software companies across
 all stages of maturity. • Additional capital may be required for the longer journey to ‘minimum viable product’ and
 the high cost of AI talent. • An imbalance between supply and demand for capital (many venture capital firms chasing 
 limited high quality opportunities) is also inflating round sizes. 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 -20% -10% 0% 10% 20% 30% 40% 50% 60% 2012 2013 2014 2015 2016 2017 2018 Fig. X: Since 2015, AI companies have raised larger investment rounds Median funding (US$m) AI premium over software Software ex AI AI AI premium 0 5 10 15 20 25 0% 10% 20% 30% 40% 50% 60% 70% 80% Seed Series A Series B Series C Fig. X: AI companies are raising larger rounds at all stages of maturity Median funding (US$m) AI premium over software Software ex AI AI AI premium 

Core technology companies attract a disproportionate share of funding Source: MMC Ventures, Beauhurst, Crunchbase, Tracxn • While core (‘deep tech’, sector-agnostic AI) technology companies comprise a tenth of AI startups, 
 they attract a fifth of venture capital investment. • The cost of world-class talent, extended time to achieve a minimum viable product, and alternative 
 revenue models (e.g. licensing agreements) can increase cash requirements. Core tech attracts a disproportionate share of funding 77 5% 10% 15% 20% 25% 30% 35% 40% 45% 50% 0% Core Technology Function Sector Fig. X: Covers companies founded since 2014 Share of funding 19% 12% 34% 41% 47% 47% Share of total (number) 

A greater proportion of AI startups are highly valued Source: Dealroom.co, MMC Ventures A greater proportion of AI startups are highly valued (detail – lower values) Source: Dealroom.co, MMC Ventures AI companies attract premium valuations 78 The State of AI 2019: Divergence • AI companies are securing higher valuations as well as more capital. • Beyond investment mechanics (larger rounds require higher valuations to avoid excessive dilution), and industry fundamentals, an imbalance in demand for capital and its supply is increasing valuations. • With AI entrepreneurship becoming mainstream, this valuation tailwind is likely to reduce. AI startups Software (ex AI) startups Proportion of companies Valuation bracket (€m) 0% 10% 20% 30% 40% 50% 60% 70% AI startups valuation data 1 0–5m 5m–10m 10m–15m 15m–20m 20m–25mm 25m–50m 50m–100m AI startups Software (ex AI) startups Proportion of companies AI startups valuation data 2 0–1.0m 1.1m–2.0m 2.1m–3.0m 3.1m–4.0m 4.1m–5.0m 0% 10% 20% 30% 40% 50% 60% Valuation bracket (€m) 

79 • Recruiting AI talent is challenging. Startups compete with multiple categories of competitors (large technology companies, banks, professional service firms and other early stage companies) for a limited pool of AI practitioners. Data, talent and productising AI are key challenges Competition for AI talent, the limited availability of training data, and the difficulty of moving AI from ‘lab to live’ are entrepreneurs’ key challenges. “Access to talent, and its competitiveness is the
 biggest challenge.” David Benigson, Signal Media “London has one of the best pools of talent
 in the world – and is the main reason we’re here.” Fabio Kuhn, Vortexa 

80 • Access to initial training data sets poses a challenge. Startups are mitigating the difficulty by drawing on a diverse range of data sources, developing powerful use cases for access to client data, and by implementing a data acquisition strategy from early in their lives. Data, talent and productising AI are key challenges “We started collecting data very early in our journey.” Timo Boldt, Gousto “It’s a classic chicken and egg problem. Early customers, and thus data, are hard to get when you don’t have any existing reference clients.” Tim Sadler, Tessian The State of AI 2019: Divergence 

81 • Developing production-ready AI difficult. Entrepreneurs recommend moving ‘from lab to live’ as soon as possible, testing development systems on low-risk real-world data. Cross-functional collaboration is key. Data, talent and productising AI are key challenges “Taking what works well in a lab and getting it to work in a diverse and sick population is a big challenge.” Chris McCann, Current Health “The real world is full of black swans and exceptions. We’ve learned to overcome them by getting great at cross-functional collaboration and constant monitoring of risk.” Timo Boldt, Gousto The State of AI 2019: Divergence 

82 • Our AI Playbook is an accessible, step-by-step guide to taking advantage of AI in your company. • A blueprint for success with best practices in AI: strategy; data; talent; development; production;
 and regulation & ethics. Our AI Playbook: your guide to taking advantage of AI The State of AI 2019: Divergence Download the Playbook free at
 mmcventures.com/research 

The State of AI 2019: Divergence 5. The implications of AI 

The implications of AI The State of AI 2019: Divergence 84 AI will have profound implications for companies and societies. AI will reshape sector value chains, enable new business models and accelerate cycles of creative destruction. While offering societies numerous benefits, AI poses risks of
 job displacement, increased inequality and the erosion of trust. 

AI offers innovation, efficacy, velocity and scalability Source: MMC Ventures AI offers innovation, efficacy, velocity and scalability… 85 

AI will disrupt companies and markets by enabling: • New market participants • Shifts in sector value chains • New business models • New commercial success factors • Changes in companies’ competitive positioning • Shifts in skills and organisational design • Accelerated cycles of innovation 86 …with significant implications for markets The State of AI 2019: Divergence 

• By automating capabilities previously delivered by human professionals, AI will
 reduce the cost and increase the scalability of services. • Markets including healthcare and financial services will see broader global participation. 87 AI will enable new market participants… The State of AI 2019: Divergence • Healthcare example: medical diagnosis is expensive for people in developed nations and inaccessible to many in developing economies. • AI-powered automated diagnosis available at negligible marginal cost, coupled with extensive smartphone penetration, will reduce the cost and increase access to primary care – enabling an influx of new
 market participants. 

• AI will change where, and the extent to which, profits are available within a value chain. • Value chains in sectors including insurance, legal services and transport will be reshaped. 88 …and drive shifts in sector value chains The State of AI 2019: Divergence • Transport example: 42% of revenue from global insurance premiums come from car insurance. As AI-powered autonomous vehicles gain adoption,
 the frequency of car accidents will reduce. • The profitability of car insurance policies will fall – by as much as 81%
 in the UK (Autonomous Research). • How will insurers adjust to a collapse in profitability of much of their revenue? • Multiple sector participants must plan for a profound shift in their sector’s value chain. 
 42% of the revenue from global insurance premiums comes 
 from car insurance. Source:
 Autonomous Research 

• AI, growth of ‘x-as-a-service’ consumption, and subscription payment models will obviate select business models and enable new ones. • Expect new business models in sectors ranging from healthcare to transport and insurance. 89 AI will enable new business models… Transport example: AI will transform the economic fabric of ownership
 and insurance. • Historically, wasteful private vehicle ownership has been required to provide spontaneity, convenience, security and privacy. • On-demand access to a fleet of autonomous vehicles will provide the same
 at a disruptive low cost. • Expect ‘transport-as-a-service’: pay a fixed, low monthly fee for all-you-can-eat access to an autonomous fleet - and reduced private vehicle ownership. • ‘Downstream’ models (of repair centres, charging centres) will also be transformed. Cars are parked for an average of 96% of their lives. Source: UITP Millennium
 Cities Database 

New technology paradigms offer significant benefits but demand new competencies. Winners in the age of AI will develop the new competitive success factors required for AI. Success factors for the age of AI include the ability to: • possess the vision to embrace AI and the organisational changes it requires • leverage large, non-public data sets to train market-leading algorithms • evaluate the opportunities and risks of sharing training data with partners and competitors • attract, develop, retain and integrate data scientists • form effective partnerships with ‘best-of-breed’ third-party AI software providers • diligence AI partners effectively • understand and respond to regulatory challenges posed by AI • effect a mindset shift to software that provides probabilistic instead of binary recommendations • manage workflow changes resulting from AI implementations • manage challenges of organisational design and culture as AI augments and replaces personnel 90 …and demand new commercial success factors 

New platforms, disruptors, leaders and laggards will emerge as AI disrupts companies’ competitive positioning: Providers of AI: • Platforms: With vast data sets, world-class AI teams and extensive resources, select GAIM vendors (Google, Amazon, IBM and Microsoft) will accrue value as platforms that support the provision of AI. GAIM lack the strategic desire or data advantage to address myriad domain-specific use cases, presenting opportunities for Disruptors. • Disruptors (early stage, AI-first software companies tackling business problems in a novel way using AI): enable SMBs and Enterprises who embrace them while eroding the value of those that do not.
 Buyers of AI (SMBs and Enterprises): • Leaders will emerge in key industries by: anticipating the shifts in value chains and business models caused by AI; taking advantage of large, proprietary data sets to train AI algorithms; having the organisational ability to deploy AI effectively; and by having sufficient resources and reputation to attract high quality AI talent. • Laggards lack the will or organisational ability to embrace AI. While some lack the foresight to adapt, more will falter due to limited organisational capability. 91 AI will cause shifts in companies’ competitive positioning… The State of AI 2019: Divergence 

• As AI gains adoption the skills that companies seek and companies’ organisational structure will change. • Currently, companies’ engagement with data scientists is limited; a mix shift to employing data scientists is likely. • Hiring for adaptability will be increasingly important, as the range of tasks supported or undertaken by AI systems increases. 92 …and drive changes in skills and organisational design The State of AI 2019: Divergence • Professional services example: global professional services and consulting firms average 5,000 to 15,000 in-house analytics professional; fewer than 8% are data scientists (MMC Ventures) . • Tomorrow’s leaders are aggressively expanding their data science teams; time to market is key given competitive advantage through data network effects. 41% of companies are considering the impact
 of AI on future skill requirements. Source: PWC 

Cycles of innovation, adoption and consumption are compressing Source: European Environment Agency, based on Kurzweil AI will accelerate cycles of innovation 93 The State of AI 2019: Divergence • By reducing the time required for process-driven work, AI will accelerate the pace of innovation. • This may compress cycles of creative destruction, reducing further the period of time for which all but a select number of super-competitors maintain value. Cycles of innovation, adoption and consumption are compressing Time before mass use Long Short Black and white television 26 Invention available to the general public 1926 1873 1897 1876 1870 1880 1890 1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 1983 1991 1991 1979 1975 1951 Computer 16 Mobile phone 13 Compact disc 12 7 10 World Wide Web Smartphones 31 Radio Telephone 35 Electricity 46 Colour television 18 Years neccessary for an invention to be used by a quarter of the US population 33 years In 1965, average company
 duration in S&P500. 14 years By 2026, est. average company
 duration in S&P500. Source: Insight 

In addition to numerous benefits, AI presents challenges and risks for societies: • AI-powered automation may displace jobs • Biased systems could increase inequality • Artificial media will undermine trust • AI offers states greater control and poses trade-offs between privacy and security • Autonomous weapons may increase conflict between nations 94 To societies, AI presents risks as well as benefits The State of AI 2019: Divergence 

• AI will directly enable the automation of select occupations that involve routine and repetition – from truck-driving (3.6 million US jobs) to telemarketing. • AI will augment and then displace workers in some more complex roles, while reducing the need for additional workers to be hired as companies expand. • However: analysis of UK census data since 1871 shows: historically, contracting employment in agriculture and manufacturing (due, in part, to automation) have been more than offset by rapid growth in alternative (caring, creative, technology and business service) sectors (Deloitte). • Greater automation of manual and business service roles may concentrate employment further in occupations resistant to automation – including care work and teaching. 95 AI-powered automation may displace jobs The State of AI 2019: Divergence • Whether or not, over time, AI creates more jobs than it destroys, the short period in which select workers could be displaced, with a reduction in the availability of similar roles, could prevent those who lose their jobs being rapidly re-absorbed into the workforce. Social dislocation, with political consequences, may result. 

AI-powered facial recognition systems misgender 1% of lighter- skinned males but 35% of darker- skinned females Source: J Buolamwini, M.I.T. Media Lab, via The New York Times 96 Biased AI systems could increase inequality The State of AI 2019: Divergence • Theoretically, AI can free decision-making from human bias
 by finding objective patterns in large data sets. • In practice, AI systems learn by processing training data.
 Available training data reflects systemic historic biases, particularly regarding gender and race. • Incomplete or inadequate training data are causing AI systems to behave problematically, particularly to minorities. • Example – in a popular data set used to train facial recognition systems,
 75% of faces are male and 80% are white. • As a result of poor training data, facial recognition systems that offer gender 
 classification misgender 1% of lighter-skinned males but up to 7% of
 lighter-skinned females (Buolamwini, Gebru). 

• AI systems are being used to make a growing range of decisions that have a significant impact
 on individuals’ lives (credit, recruitment). •“If we fail to make ethical and inclusive AI, we risk losing gains made in civil rights and
 gender equity under the guise of machine neutrality.” Joy Buolamwini There are potential harms from algorithmic decision-making Source: Megan Smith via gendershades.org 97 Biased AI systems could increase inequality INDIVIDUAL HARMS HIRING EMPLOYMENT INSURANCE & SOCIAL BENEFITS HOUSING EDUCATION ILLEGAL DISCRIMINATION UNFAIR PRACTICES COLLECTIVE SOCIAL HARMS CREDIT DIFFERENTIAL PRICES OF GOODS ECONOMIC LOSS LOSS OF LIBERTY INCREASED SURVEILLANCE SOCIAL STIGMATISATION STEREOTYPE REINFORCEMENT DIGNITARY HARMS LOSS OF OPPORTUNITY Potential harms from algorithmic decision-making 

98 Biased AI systems could increase inequality The State of AI 2019: Divergence To avoid ‘automating inequality’ developers can: • recognise the challenge, as a starting point for action; • develop diverse teams that reflect the communities they serve; • create balanced, representative data sets; • deploy rigorous ethics and testing frameworks for system validation. There is a battle going on for fairness, inclusion and justice in the digital world.”
 (Darren Walker, via The New York Times). 

Given video of former President Obama, researchers synthesised photorealistic , new lip-synched video Source: Suwajanakorn, Seitz and Kemelmacher-Shlizerman • Generative Adversarial Networks (‘GANs’) are an emerging AI technique that enable the creation of
 artificial media, including pictures and video, impossible to differentiate from real content. • GANs will democratise the creation of content, enabling the production of media at scale and low cost. • GANs also enable the creation of damaging counterfeit media in which individuals appear to perform actions they have not undertake or speak alternative dialogue. 99 Artificial media may undermine trust – ‘fake news 2.0’ • In addition to damaging individuals, artificial media will undermine trust. If any media can be counterfeit, all media is open to challenge. • In the age of artificial media, society
 will grapple with challenges of truth and trust. • Adversaries recognise that sowing doubt and confusion to divide populations can be more powerful than direct action. 

• AI-powered facial recognition has unprecedented capability. To what extent will citizens and governments sacrifice anonymity to prevent crime? • AI with real-time analytics are enabling the high-tech surveillance state with greater capacity for social control. China intends to combine real- time recognition with social scoring to inhibit ‘undesirable' behaviour. 100 AI presents new trade-offs and offers greater state control The State of AI 2019: Divergence “You don’t need people’s cooperation for us to be able to recognise
 their identity” (Huang Yongzhen, Watrix, via the Associated Press). 
 A citizen is captured on
 CCTV an average of 68 times per day. Source: Cheshire Constabulary Camera Survey 

• AI-powered computer vision systems, AI-based decision-making algorithms and improved robotics are enabling humanoid and aerial drones with greater capability and autonomy. • While the risk of ‘killer robots’ turning against their masters is overstated, less considered is the possibility that conflict between nations may increase if the human costs of war are lower. 101 Autonomous weapons may increase conflict The State of AI 2019: Divergence A country that thinks twice about sending young people into conflict may be more adventurous if the only assets in harm’s way are equipment. 

MMC Ventures is a research-led venture capital firm. We’ve backed over 60 early-stage, high growth technology companies since 2000. AI is a core area of research and investment.
 We’ve made 20 investments into many of the UK’s
 most promising AI companies. If you’re an early stage AI company, get in touch
 to see how we can accelerate your journey. www.stateofai2019.com Get in touch 102 MMC Ventures Research David Kelnar: Partner & Head of Research
 Asen Kostadinov, CFA: Research Manager Explore MMC’s cutting-edge research at mmcventures.com/research. 

No content