Ep. 516 Bob Murphy Interviewed About the Future of AI
Bob recently appeared on the “Human Reaction” podcast, where host David Rand asked him various questions concerning the economics of AI. Although frequent BMS listeners will have heard these themes before, there is new material in this discussion.
Mentioned in the Episode and Other Links of Interest:
- The YouTube version of this episode.
- The Human Reaction YT channel.
- This episode’s sponsor, the free Plan-B guide from ExPatMoney.
- Help support the Bob Murphy Show.
About the author, Robert
Christian and economist, Chief Economist at infineo, and Senior Fellow with the Mises Institute.
I think you are dramatically overestimating the productivity gains AI will be able to deliver. You said something about the best workers getting their output doubled or tripled, but there is no chance of that happening. The reason for this is that AI is ultimately just a faster search engine. But the top workers have already mastered search within their domain. Most have the most valuable things already committed to memory, so using AI won’t help at all for those cases. They also know the authoritative materials and how to navigate them for situations that they don’t have memorized – and again using this method is faster (and cheaper) than AI. AI is only going to deliver significant gains when stepping outside the domain of their expertise and hiring a whole human expert for that domain is overkill.
It was also mentioned a couple times that the AI can be used for legal questions. This is an EXTREMELY bad idea. For one, the AI cannot be held accountable for bad answers, while an actual lawyer can. For another, there is no attorney-client privilege for your questions to ChatGPT. If you get sued, every prompt you gave will be brought up in court and will only be used against you.
“The reason for this is that AI is ultimately just a faster search engine.”
You have no idea what you’re talking about, unless you are speaking in some tautologous sense in which Albert Einstein working out general relativity was just searching the space of possible theories of gravity.
Actually I know perfectly well what I’m talking about, because I know how these things work. They are built from the data scrubbed from the internet and other published works, and they just probabilistically search through that data in order to build responses. The reason you are so impressed by the code they write is because you are not a programmer.
It’s not just a faster search engine, but it is primarily a faster search engine. However, there’s some key points here … it’s way easier for me to go forwards and backwards in a “chat” style format and get neatly summarized explanations, than it is for me to read through 10 or even 100 search results and put together my own summary of all that information. Search engines themselves were already way easier than the way we used to do it, which was going to a library, working through the catalog cards, pulling out books then taking a photocopy of a few pages … then hole punching and clipping pages into binders.
Doing the same job better … is still better … especially when it’s a job many people want to do. Because it’s better, users steadily move over and give up on the old way of doing it.
I mean, a mobile phone is just a phone, but more convenient … and yet, because of that little bit of convenience most people do prefer the mobile version of the phone to the old version which required you to sit on that little bench in the hallway.
The Internet is primarily doing the same job as what a FAX machine used to do. We used to shuffle through a paper catalog, then fill in the purchase order and FAX it off, and the guy at the other end would respond and get your stuff ready for shipment. We used to put mail in a paper envelope and walk it to the post box. Now we send emails, or we do the same via “blogs” or social media, and when we want to buy something we use an online catalog like Amazon, eBay, DigiKey, etc … and fill in a form by clicking the part instead of copying the part number. It’s the same process, but better, more streamlined.
This of course gets into the question of profitability. Not many companies made huge profits on selling “Internet” because it rapidly became a commodity. Sure you have Comcast, they do OK, but not spectacularly. The fact that many people shifted to the better method of doing things, is no guarantee that any particular business is the next big winner. Same with AI … I think people will be using it, but the software and hardware will get more efficient, the whole thing gets commoditized, and pretty soon profit margins are razor thin. If you were smart enough to think ahead and buy Nvidia shares back in 2021 then because you got in early you could have made huge gains … but there’s nothing smart about getting in after the discovery has happened.
I’m not saying that AI tools can’t increase productivity, just that the numbers Bob gave are wildly unrealistic. You aren’t just reaping free gains because your time is now spent verifying the output instead of creating it yourself, and this takes a lot longer than the people selling you these AI tools are pretending it does.
And the accountability problem in my original post still remains. This is not only a thing in the legal industry. If you write software for hospitals, YOU are accountable when your software fails and people die. If you write financial software, YOU are responsible when your customers get wiped out or have their data breached. You don’t get to say “b-b-b-b-but the AI did it!” AI cannot be held accountable.