Diplostack #10: AI cool sh*t, holy sh*t, or bull sh*t (1/2)
From healthcare to agriculture, AI has the potential to revolutionise the way we work and live but let us keep the fun stuff for us humans ...
Cool/holy/bull-sh*t -- bit of everything it seems; hear from a range of practitioners on their views on AI in their fields: agtech & foodtech, software development, healthcare, special FX/movies, the environment, and business productivity, with: the Prime Minister's Chief Science Advisor, Sprout Agritech, MIT, Github, Weta FX, Wellington Hospital Spine Surgeon, and a leading NZ AI startup founder.
Hosted by Maty Nikkhou-O’Brien, founder & Executive Director of Diplosphere, this was the first panel session recorded in Wellington on Aug 29, 2023 at Diplosphere’s Conference 2023 on AI.
Listen to the whole podcast here:
Chair: Professor Dame Juliet Gerrard, Prime Minister’s Chief Science Advisor
Speakers:
(00:00) Introduction by the NZ Prime Minister's Chief Science Advisor
(06:38) Gil Meron CEO at Sprout Agritech
(13:13) Priya L. Donti Associate Professor at Massachusetts Institute of Technology and founder of Climate Change AI
(20:29) Igor Costa Senior Architect at GitHub
(26:14) Kimball Thurston CTO at Wētā FX
(33:25) Dr Shay Mandler Spine Surgeon at Wellington Regional Hospital
(41:37) Asa Cox Founder and CEO at Arcanum AI
First three summaries below (AI assisted from human bullet point notes).
But before we start, don’t miss Diplosphere’s next event on Thu 2 Nov in Wellington: God Defend New Zealand - From What? On New Zealand's independent foreign policy and the geopolitical challenges an incoming government will need to navigate.
Food & Agri-tech - Gil Meron @ Sprout Agritech
1/ In this section, Gil Meron, CEO at Sprout Agritech, describes the transformative potential of AI in agriculture, ranging from data analytics to automation. [editor’s note: it appears we are in “moment” in foodtech & agtech with an explosion of data, connectivity to rural almost solved, and now the AI/analytics to make sense of it all]
2/ Top decile farmers - Digital assistants like Aimer, a Sprout investment portfolio company, can help turn average farmers into top decile farmers. Technologies like Generative AI (GenAI for short) offer specialized advice to farmers, and big data analytics help in expediting processes related to agrichemicals and genetic breeding. We are great in New Zealand in breeding animals, and selecting plants, but imagine what we could do if we could speed up that process by scouting all the genetic data available? Vast amount of genetic data sit untapped, today.
3/ Bio-sensors - Innovative approaches are emerging, such as using plants themselves as 'living sensors' to detect heat stress or diseases, which could revolutionize monitoring and treatment in agriculture.
“I have heat stress, or I have a disease”. -Plants of the future
4/ The dog is the gold standard — Sprout backed synthetic nose company Sentient Bio with its AI-enabled 'sense of smell' on a chip, standing out as Gil's biggest Holy Sh*t Moment. This technology, the equivalent of a nose on an insect, can identify a variety of scents 1000x more sensitively than dogs which are in turn 1000x more sensitive to smell than humans. This can be used to smell residue of chemicals on our food, to sense disease (the Gates Foundation is financing the company to detect tuberculosis), and other applications. This is technology that has been around for many years, but enabled by AI. Electric and autonomous vehicles are also becoming much more prominent in agriculture. Think autonomous tractors, or swarms of drones (example: NZ Zero’s Mike Casey and his Monarch tractor, a first)
5/ The global food system, traditionally optimized for cost, faces a paradigm shift. In light of the COVID-19 pandemic and technological advances, there's an increasing focus on optimizing for food security, health, and other essential elements.
The global food system has been optimised for cost for decades; with Covid pandemic, and AI advances, opportunity with AI to optimize for food security, for health, for everyhing we need our food system to give us
Climate & AI - Priya Donti @ MIT
1/ Priya Donti, incoming Associate Professor at MIT and co-founder & Executive Director of Climate Change AI. At MIT, Priya’s research focuses improving algorithms for power grid optimisation in order to foster the integration of renewables. And non-profit Climate Change AI brings together people working on Climate Change and AI aiming to provide education, resources, networking, funding.
AI represents huge opportunities to accelerate our ability to both reduce greenhouse gas emissions and adapt to the effects of a changing planet when it is applied responsibly and partnershp with relevant stakeholders
2/ Smarter energy usage and deterring 192 billion locusts - On the one hand, Priya Donti highlights positives like how AI can: improve predictions & forecasts, distill large unstructured datasources into actionable information, optimise complex real world systems, and accelerate scientific discovery. Priya provided examples to illustrate each case. AI can drastically improve predictive analytics, illustrated by the UK National Grid ESO's success with the help of Open Climate Fix in reducing electricity demand forecast errors by 50% by using deep learning on historical data and weather data - critical to balancing power grids with large amounts of renewables (better solar forecasts could reduce UK emissions by about 100,000 tonnes per year). Beyond just predictive capabilities, AI has the power to make sense of vast, unstructured data sets. This is seen in projects like MAPP, which allows real-time monitoring of Amazon deforestation, and the UN Satellite Centre's high frequency flood monitoring in Africa and Asia. Moreover, AI is also capable of optimising complex systems, like Arup Hong Kong’s work in enhancing building energy efficiency (improves energy efficiency 10-30%), and expediting scientific research, such as AI Onics' efforts in cutting down battery design time. AI’s predictive capabilities can also be used for environmental disasters like locust outbreaks.
Swarms containing up to 192,000,000,000 locusts – covering an area three times the size of New York city – were spotted in Kenya in 2020. A locust swarm of this size is estimated to consume the same as 90,000,000 people in just one day.
Locust early detection system, Instadeep, 2021
3/ AI is No Silver Bullet for Climate - On the other hand, AI's role in climate action isn't without drawbacks. Somebody has to act on the predictions or optimisations from AI. AI is - at present at least - is not going to make hard policy decisions. Value judgements are embedded in these AI systems. And values and judgements are a function of a particular society. At the same time, development of AI tech is presently done by a small number of well funded organisations (like big tech firms) - many of whom are shaping agenda on how AI gets used, and perhaps not considering value judgements of particular societies. AI is also used in detrimental ways for climate action like: improving oil & gas extraction processes, driving consumption through personalised advertising, and AI has its own energy footprint from compute, its hardware, power, and water usage for DCs.
AI is a key technology in targeted advertising which invariably increases how much we consume, which increases our footprint, without necessarily making us happier
4/ It’s a balancing act - There are lots of ways AI can help address climate change related problems, also need to make sure we are being holistic in how as we as a society shape the usage of AI so that is is aligned with our environmental and other social goals
Creativity & AI - Kimball Thurston @ Wētā FX
1/ Kimball Thurston CTO at Wētā FX started by showing the room all the cool sh*t they have done over the years. It served as a primer for discussing the complexities around data provenance, touching on questions of data rights, ownership, and authenticity.
2/ Rights & privacy: who has the right to use what data? - There's a critical need to protect proprietary data, especially in a creative industry where intellectual property like characters from the 'Avatar' movie can't cross over into another movie house’s production. This sensitivity extends from customer data - the show, the scene, the shot - down to specific scenes and internal data - generated internally, and workflow data - how Wētā FX makes itself more efficient.
3/ Ownership: who owns the data? - Humans have their own creativity. A person who creates something, may be able to copyright it and can assign it to someone if they want to sell that. Questions arise about who owns the output of machine learning algorithms, adding another layer of complexity to the intellectual property discourse.
Is the output of an ML algorithm "ownable"?
4/ Authenticity - Is the data real or not? Thurston also touched on the ethical dimensions of technology, notably the potential risks of malicious intent with the state of the art of the technology today. The known techniques used to create a blue character on an alien planet are not so different from those used to produce a fake video. While initiatives like Adobe's C2PA aim to authenticate data, they pose challenges in terms of remaining agile. Weta FX is actively contributing to a file format, to improve data authentication methods without sacrificing efficiency. [ed: is this Web3’s moment in combination with AI? see Connective Tissue by David Levy]
5/ Finally, Thurston emphasised Wētā FX's ethos as creators, not mere tool builders. Although machine learning offers many promises, it doesn't necessarily make processes faster, cheaper, and better simultaneously. The company focuses on creating tools that augment human ability rather than replace it, with the ultimate goal of keeping the 'fun' in the creative process.
Let's make sure we're talking about creating tools, not replacements for people. We need a diverse thing. Language models are just language.
Tap the Like button, and use the Share button to share across social media. And feel free to leave a comment if you are so inclined, and remember to subscribe if you haven't done so already.