I loved this interview with author Steven Johnson. It covers a lot about the way he writes and how he structures his books and articles. It also reminded me about the piece he wrote back in 2022 on ChatGPT which has a wonderful opening paragraph that illustrates the simple premise behind Large Language Models – predicting the next word.
The article still rings true even though the technology has moved on a lot. Re-reading it reinforced my view that what people are calling ‘AI’ is really just ‘applied statistics’ (as Ted Chiang calls it. That doesn’t make it any less interesting but it does put some boundaries around how we should think about it.
I’ve been working around the edges of machine learning and AI for many years now. We implemented some basic machine learning in the startup I was running in 2006 and I watched as many other startups implemented similar things during the 2010s. As an investor at BGV I’ve seen our portfolio companies use AI to differing extents to build successful businesses and have a positive impact.
So I knew that the recent spate of LLM innovation was coming. It’s impressive to see what companies like OpenAI and others have achieved. There’s something uncanny about the interactions you have with ChatGPT or Bard and the like and I’ve watched it already have an impact on the stuff I read.
Unfortunately, the most obvious thing in my case is inbound unsolicited marketing emails. As I’m fairly public about my contact details, I’ve always had a fair amount of speculative sales messages from recruiters, outsourced software development houses, lead generation and many other services that I just never use. It doesn’t get flagged as spam but I use software called Sanebox (itself an interesting application of machine learning) to filter it and then have a quick scan once a day.
In the last couple of months, the nature of those emails has changed. They are now mainly generated by chatGPT and the like. Because there is enough information about me and BGV in the public domain, they can ‘personalise’ the approaches in a way that wasn’t possible. I’ve also noticed a fair amount of ChatGPT generated posts on Linkedin, Twitter and the like. It’s an interesting twist to sales and marketing but it leaves me underwhelmed.
Despite how impressive the technology is, so far I’ve found limited use cases for ChatGPT in ‘doing’ any part of my job. I find it useful for sense checking and improving the quality of output but it’s not capable of fundraising or making investment decisions. It can help but it’s a long way from being a direct replacement for human activity.
I’ve been asked quite a few times in the last six month about the relationship between tech for good and AI. The short answer is that it’s no different from any other technology. A tech for good AI startup will set out to intentionally solve a particular social or environmental issue and it will measure its impact as it tries to do that. No other AI startup will be tech for good. You can’t accidentally be tech for good.
Part of the reason for this post is that I think ChatGPT and the like will lead to people writing less and that is a shame. Seeing it in action has spurred me to do something which I’ve been thinking about for a while and start blogging again.
I’m going to try to write a weekly post, usually about tech or impact investing. Maybe I’ll just be talking to the bots. Does anybody ready blogs anymore? I’m not sure!
Everything on the internet is wrong so don’t take me too seriously. But I hope it will help me improve my thinking which is something that leaving everything to AI certainly won’t.
I’m a terrible Go player. Perhaps that’s why I hadn’t quite understood AlphaGo until recently reading more about it in Erik Brynjolfsson and Andrew McAfee’s book Machine, Platform, Crowd.
Machines became better than humans at chess a while back as increasing computing power enabled Deep Blue and the like to calculate possible moves and rank whether they were likely to help the computer win. But there are more possible positions in Go than there are atoms in the universe. In fact there are enough possible positions for there to be a universe of atoms for every atom in the universe (that’s 10 to the power 82 in case you’re wondering).
Even today cracking that by brute force would require more computing power than we have available. But what I hadn’t understood was that the best human Go players don’t know why they’re so good.
“How do the top human Go players navigate this absurd complexity and make smart moves? Nobody knows — not even the players themselves. Go players learn a group of heuristics and tend to follow them. Beyond these rules of thumb, however top players are often at a loss to explain their own strategies. As Michael Redmond, one of the few Westerners to reach the game’s highest rank, explains, “I’ll see and move and be sure it’s the right one, bit won’t be able to tell you exactly how I know. I just see it.”
So programming a computer to play Go is tricky because we don’t know what to teach it. What AlphaGo did was teach itself. Back in the 1950s when the theoretical basis for artificial intelligence was being laid out there were two branches — one was rule-based (a bit like the way adults learn a new language) while the other was essentially statistical (like a child learning to talk by trying things over and over again). What AlphaGo shows is just how powerful this second version of AI has now become.
I signed up to the waiting list about a year ago. I was visiting Craig and Kanyi at the Collaborative Fund in New York and Craig had simply copied ‘Amy’ into our emails to sort out the details. I noticed the strange email address straight away and did some digging. Amy it turned out was an artificial assistant created by New York based company x.ai. The process worked without a hitch and I remember thinking how unremarkable the whole experience was.
Even though I knew Amy was an AI, I remember feeling that I should be polite to her. I generally make a point of thanking all the PAs who arrange meetings with other people for me and treat them in the way I would treat the person I’m meeting. It was hard to shake that habit.
Fast forward nine months and I got the email saying that I could now use Amy too as the beta trial grew. I’d actually forgotten I’d signed up (nobody else I’d met had been using x.ai in the period between), but I quickly clicked on the link and went through the very simple setup process of sharing calendars and answering some basic questions about common types of meeting (how long they should be, favourite locations etc). I also got to choose whether I wanted Amy or Andrew and decided on Andrew.
Then I got a bit stuck. It was not so much that I’d never had an AI personal assistant before — I’d never had a personal assistant at all.
Before trying it out on others I felt I needed to give it a quick test so sent my colleague Vicky at BGV an email suggesting we go for a coffee and that Andrew (cc’d) could arrange a time. Now Vicky in a past life used to be an executive assistant so I thought she might be intrigued. In fact she was just rude and deliberately awkward making Andrew’s task all the more difficult by changing her mind and suggesting venues that she knew wouldn’t work. Andrew gave up and politely ‘reverted this one back to me’.
Next I tried letting him organise a few phone calls that weren’t time sensitive and Andrew did fine. Then a few coffees, which also went fine. I didn’t ever use it for my social life — that would seem a bit weird to me — and if it was a really important meeting I still did it myself.
I did a bit of maths and realised scheduling takes about 5-10% of my working day so anything that can reduce that is very valuable. It’s also a real drag — even the PAs I know would rather do more valuable things if they could so my feeling at the moment is that x.ai is creating job displacement rather than job replacement.
Most people I meet who have interacted with Andrew want to talk about it and they usually only have positive things to say. A couple of people didn’t notice that he’s an AI at all and a lot of people have asked if I can get them bumped up the waiting list (I can’t).
I’m not sure whether I’ll go all in and let Andrew organise all my meetings. It’s going to take a bit of getting used to but the barriers are more human and social (what other people think) than technological. It’s not quite ‘Her’ or ‘Ex Machina’ but it does feel like the future, albeit in a very mundane way.