This as-told-to essay is based on a conversation with William Tunstall-Pedoe, a 57-year-old founder and CEO who sold his voice recognition company Evi Technologies to Amazon in 2012, becoming a co-creator of Alexa.
Business Insider previously interviewed Tunstall-Pedoe about how Evi, an AI-powered search engine that answered spoken or typed questions, became an integral part of Amazon’s Alexa. He told us his addiction to tech started in his high school and that early in his career, after graduating from Cambridge in 1991, he worked for Acorn Computers and the Isaac Newton Institute, where he did what he called “some more niche AI things.” After leaving Amazon in 2016 to mentor founders and invest in AI startups, he launched UnlikelyAI, a neurosymbolic AI startup, in 2019.
Tunstall-Pedoe weighed in on OpenAI CEO Sam Altman’s recent claim on the “Relentless” podcast that “we are now, like, in the singularity” — the point at which AI surpasses human intelligence and begins advancing faster than humans can predict or control.
This essay is part of Business Insider’s coverage of expert responses to Altman’s claim. It has been edited for length and clarity.
All sorts of friends of mine who aren’t in technology have been reaching out to me over the last few weeks because they’re suddenly having to wake up to what’s happening at work or in their industry due to AI.
The number of people using LLM chatbots for all sorts of things is only growing, and people are becoming more aware of this change. Still, there are a lot of assumptions about AI singularity and what it means.
To me, the singularity is the idea that technology accelerates improvements in technology at an ever-faster rate. Eventually, it’ll get to the point where the speed of technological change is so enormous and autonomous that the human race will get caught up in it and land in some magical future state.
AI is certainly continuing the trend of new technology accelerating old technology, but that’s been happening for centuries, with the Industrial Revolution being a prime example. As we’ve seen in the past, it’s not a new and novel idea for a machine to be better than humans at a task.
I think the idea of runaway AI, brilliant and perfect at everything and at a level of intelligence that’s thousands of times ahead of what we see today, is implausible at the moment. I don’t think we’re in the singularity.
Machines have outperformed us in some tasks for decades
If you look at what the brain can do, there are things that computers have been able to do dramatically better than the brain for decades and decades.
If you think about math, people long ago stopped trying to compete with computers at math. If you think about memory, Google has perfectly learned hundreds of billions of webpages and can produce them on demand. Meanwhile, the human brain struggles to remember six or seven digits a lot of the time.
In some dimensions, LLMs are now superior to the human brain. They can talk competently on almost every academic subject, which no human being can do. They’ve learned competencies from large amounts of training data that cover a wide range of tasks and topics.
My point is, there are ways in which machines have been dramatically superior to the human brain for a long time; this isn’t something brand new to worry about.
AI functions on human data
Part of the illusion of LLMs’ extraordinary capabilities and why they appear so smart is that they’ve been trained on a meaningful percentage of everything that’s ever been written; they’ve seen, read, and learned so much from training data.
But much of what we’re seeing is based on the imitative capabilities that come from AI training on large amounts of data and a wide range of human actions. I don’t think AI singularity is a natural continuation of that.
AI would need to show stronger evidence of what I call discovery intelligence: AI independently producing and evaluating genuinely valuable, paradigm-shifting ideas. LLMs can currently generate novel ideas, but they can’t determine whether those ideas are realistically super valuable.
The singularity idea, to me, assumes that discovery-type intelligence is an extension of the learned-competence type of intelligence that AI already has.
Here’s what I think we’d see if AI reached the singularity moment
If AI ever gets good enough to self-improve, actually innovate in AI, and it turns out there are no limits to its intelligence, then that would clearly be a very momentous point in time.
AI would then become extraordinarily capable very, very quickly because it will be able to keep improving itself in inventive ways. That rapid pace will be the singularity moment.
Generally, I think it’s easy to critique the idea of the AI singularity because it kind of assumes reality is like a mathematical function and that intelligence is very linear. I think intelligence is a collection of things that you can do to various degrees of capability, so it’s not clear to me that there is such a thing as runaway infinite intelligence, and I don’t think true AI singularity level acceleration will happen.
We would recognize it if it were here
I don’t think technology necessarily accelerates very smoothly. It accelerates for a bit, hits a wall, and then there’s a decade or two before the next paradigm shift or the next big innovation unlocks further innovation.
I think even non-tech people would notice the difference between the rapid acceleration in adoption happening now and what would happen if the AI singularity ever comes. It would show up in pretty much everything.
Read the full article here
