The AI revolution is unfolding before our eyes, and there is great excitement about how its magical powers will change our lives, for better or worse. Stuart Russell, a computer science professor at the University of California, Berkeley, warns of the dangers of this technology and the need for strong safety and ethical guardrails in the ongoing development process.
One of the leading researchers in AI, Russell’s research spans nearly every area of theoretical AI. His book Artificial Intelligence: A Modern Approach (co-authored with Peter Norvig) is a standard text in universities around the world. He founded the International Association for Safe and Ethical AI and is one of the most influential voices on AI safety and ethics.
In the second of a two-part series, Russell talks about the potential emergence of artificial general intelligence (AGI) and the current state of AI research in India. he spoke Amitabh Sinha. you can read Click here for the first part of the interview.
question: When large-scale language models like ChatGPT first appeared, there was a lot of excitement. However, subsequent releases felt much the same, albeit more powerful. What will be the next big thing in the development of AI?
Russell: If I knew the answer, I would be much richer than I am now. Since ChatGPT arrived in November 2022, we have already taken one big step (maybe 1.5 steps). ChatGPT was GPT-3.5. GPT-4 was then released in March 2023, which was a huge step forward. They were able to reason about things that were clearly not present in the training data, come up with new formulations, and solve difficult problems. That was a step.
The other half-step is what we call agent AI. In the context of large language models, this means that the system can perform actions in the real world, not just print text on the screen. Not only can you send an email, tell someone to log on to a website and make a trade, post on social media, or buy or sell stocks on the stock market, you can actually send an email.
The “next thing” that major AI companies are investing heavily in is artificial general intelligence (AGI), or AI systems that can match or exceed human capabilities in all areas. Such a system not only makes you a world champion in chess, Go, and poker, but also excels in writing scientific papers, solving problems, conducting research, creating laws, diagnosing diseases, and writing poems and love letters. It would essentially be a superhuman brain.
Story continues below this ad
I often hear that it is very close to AGI. Are we?
That’s a big question. The scale of the investment was enormous, 50 to 100 times larger than the Manhattan Project, making it the largest technology project ever undertaken by humanity.
But I don’t actually expect it to be successful. My current estimate is that the chance of failure is about 75% due to current technical limitations. On the other hand, once that limitation is overcome, it becomes even scarier. The myths of every culture contain examples of creation over which we have no control. AI that we cannot control puts it in direct conflict with us. Just as I would lose when playing chess with the AI, so too would we lose in that conflict. I basically get wiped out. It wipes the floor with me on the chessboard, and it wipes the floor with humanity in the real world.
Is the evolution to AGI inevitable? Can I change my course?
Story continues below this ad
We have indeed decided not to follow the path that previously seemed inevitable. Through cloning, we moved from single-celled animals to sheep, but we decided not to clone humans because it would be so harmful.
Is nuclear technology another example?
Yes, nuclear is very restrictive. We also phased out fluorocarbons because they destroy the ozone layer. However, the upside value of AGI is immense. “Back-of-the-envelope” calculations suggest that global GDP could increase tenfold. This possibility acts as a “very strong magnet” that draws us toward the future. As for inevitability, I think there will be another 10 years of a “mini-ice age” in AI research if the current push fails.
The idea that AGI will eventually be developed is based on the premise that the human brain essentially functions like a computing machine. Has your belief in that premise changed over the decades?
In the 50 years I’ve worked in AI, my understanding of the human mind has grown, but nothing convinced me to look elsewhere. In neuroscience, there is nothing to deny the theory that what matters is the signals that pass between neurons. Piecing together the story of functions such as hearing, vision, and motor control is consistent with a 100-year-old view of signals and neurons.
Story continues below this ad
It would be a huge shock to the scientific and mathematical establishment if something completely different was happening in the brain. It is foolish to think that there is no arrangement of atoms in the universe that can do a better job than the human brain. It would be incredibly arrogant to claim that.
What is your assessment of India’s AI research scene? Should the country develop its own large-scale language models (LLMs)?
There are several considerations. One is data bias. Training data is often appropriate for specific markets, such as OECD countries, and most of it is written in English by Westerners. If you don’t have a huge model of your own, you risk falling behind, but it’s not so clear whether your business needs a unique model.
There is also economic significance. Many people fear that they will be left behind without a huge model of their own, but it is not so clear whether business applications need a unique model.
Story continues below this ad
I would rather focus on training people in universal, fundamental, and lasting skills, such as data analysis, statistics, the mathematics of machine learning, and the mathematics of reasoning and decision making.
What would you advise Indian officials when they asked you for advice on how to move forward with this idea?
My understanding is that they are not committed to the path of trying to create AGI. They are interested in creating purpose-built systems that provide value in healthcare, education, engineering, and construction, which I think makes sense. It’s also not clear whether you need to develop your own extensive language model functionality.
I think the most valuable contribution that AI has made to the world in recent years is a system called AlphaFold, a remarkable achievement that won the Nobel Prize in Chemistry last year. AlphaFold is a system that predicts protein structures from molecular sequences and is of great value for all types of biology, pharmacology, and biotechnology.
Story continues below this ad
AI makes it possible, so why haven’t we developed more products like AlphaFold?
Because we spend trillions of dollars on LLMs. Although LLM is “fancy,” practical and useful systems like AlphaFold, which won the Nobel Prize in Chemistry for predicting protein structure, do not receive the same priority.
Similarly, machine learning has improved weather forecasting using mathematically sophisticated technology unrelated to LLM. We need more stuff like this. You can’t just throw weather data into a large language model and hope for the best.
Is it better for India to “leapfrog” in technology by focusing on these utility-based products and integrating safety and ethical standards?
Story continues below this ad
Absolutely; I believe that is the correct strategy. In the long term, various futures are possible. Areas where humans still exist either have safe AI or no AI. There is no future for humans and dangerous superintelligent AI. Investing in secure AI is the only logical path forward. Because ultimately, people won’t use AI if it’s not safe to use.
