The Long Arc of Sustainable AI For Humanity Points To Open Source Models

About open source AI solutions for the benefit of humanity

Transcript

1. The Long Arc of Sustainable AI For Humanity Points To Open Source Models Harish Pillay h.pillay@ieee.org
2. Imagine an 8 year old holding these two devices
3. Raspberry Pi 5 Smart Phone
4. And the 8 yr old asks: “Can I see how it works?”
5. What would you say?
6. What’s “Sustainable AI”?
7. Sustainable AI - 2 branches Sustainability for AI Shrinking the footprint of AI systems themselves. • • • • Energy: the massive electricity demand of training large models Carbon: emissions generated by that energy Water: consumed for data-centre cooling E-waste: growing piles of specialised hardware, now obsoleted AI for Sustainability Applying AI to broader global challenges. • • • • Energy grids: optimising renewable integration Precision agriculture: cutting resource use Deforestation: monitoring via satellite imagery UN SDGs: accelerating progress toward the goals
8. Three Pillars of Sustainability • • • • Environmental Social Economic Planetary footprint People & equity Cost & resilience Carbon footprint Energy use Water consumption Material extraction • • • • Labour-market disruption Bias & surveillance Equitable access Distribution of benefits & harms • • • Cost efficiency Long-term resilience Incentive alignment
9. Environmental - GW & H20 Remember Singapore’s slogan “every drop counts”? Aren’t the DCs included in this? We have 85 DCs in Singapore 1.4GW used by operational DCs in Singapore Run models locally as much a possible. Just because there is external AI available, doesn’t mean that you should use it. Hand painted H2O droplet by my niece, Verina, 8. https://www.pub.gov.sg/Resources/Publications/WaterConservation https://www.datacentermap.com/singapore/singapore/ https://www.pinsentmasons.com/out-law/analysis/singapore-malaysia-d ata-centre-strategies
10. Social - Systems Thinking • Honest and frank relook at how organizations are run using the Systems Thinking framework • Just because you could use AI, should you? • Uncoordinated deployment of AI will lead to eventual failure (95%) and loss of trust and confidence • Seek Systems Thinking expertise NOW! https://www.sciencedirect.com/science/article/pii/S1877050915002860 https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
11. Economic - Tokenomics • Tokens are the fungible accounting stakes of AI • Locally run models reduce token bean counting • A view of all the tokens consumed in an org is needed, and needed like YESTERDAY • The usual suspects will not make token accounting easier for orgs - it is NOT in their business interests • Tokamak.sh offers an open source, self-hostable and enterprise deployable solution with governance.
12. Open Source Benefits You • • • • • • TCP/IP, ARPAnet –> Internet, 1990 Linux, 1991 - runs the Internet, Cloud and AI The Web, 1993 Python - from 1990s PyTorch, Transformers etc - from 2010s Open weight and open source models - from 2020s
13. Start With The Download Button The real question is what you are allowed, and able, to do afterwards. With open source, the universe is the limit!
14. Objections to open?!?? • Yes, open models can be modified “abliteration” (portmanteau of “ablation” and “obliteration”), and that’s a good thing. • Fear factor, some justified, but applies to closed models equally. • Meaningful openness offers a way to study, adapt, use and share and is congruent with open source values.
15. NOT a Red pill vs Blue pill debate
16. Open Unlocks The World’s Potential Harish Pillay h.pillay@ieee.org