'AI is being driven by marketing teams and not technology,' Duke University professor

AI of course has been the buzzword of 2023 and it was back in the headlines this week to name just a few Sergey Brin has returned to Google the White House says open Ai and Google have pledged to water mate Mark AI content for safety and apple is developing its own artificial intelligence systems and now according to the White House some of the big names Amazon Google meta have entered a new agreement voluntarily that will manage the risks associated with AI so there's a lot going on but is the hype the positive chatter taking a downturn our next guest thinks it might be let's bring in Duke professor and the former Chief Innovation officer for the FDIC Sultan Meg G Sultan it's good to see you again you know this year has been quite a ride when it comes to Ai and there has been certainly in the market a lot of enthusiasm around it where do you think we are in the AI hype cycle well it's great to be back and that's a fantastic question because so much of AI right now is just driven by marketing teams and not actually by Technologies and so I think it's not surprising at all that you know we're talking about the hype cycle so much has been announced there's so many you know glittering logos and and great press releases and arguments on social media and things like that but we haven't actually seen a lot of work being done and you know the current generation of AIS are really going to start having impacts over the next few years where they you know do things like streamline back office processes and things like that but that's a completely separate set of activities from uh from what gets covered by the news Sultan so okay yes AI is supposed to take on more roles within more businesses whether it's streamlining or being an assistant in news apparently uh whatever that means but there's been some reporting that AI is not as intelligent as it used to be as you layer in this you know kind of machine learning learning from itself and this kind of feeding frenzy on itself making it Dumber have you seen that in any of your research uh absolutely you know pretty much all of the large language models out there are showing signs of of Investments the wrong side of reinforcement learning so in AI you have to be really careful with the data that you use to train it and at some point it will you know kind of stop having positive returns it'll stop getting smarter and you need to stop and kind of take a step back at that point and really be more focused on the kinds of data you're analyzing or just say you know what for now this is the best we're going to be able to do and you know we're starting to see that I think pretty much every single major tech companies openly available llm is showing some sign of of plateauing in terms of its capabilities according to Summer report telling but getting Dumber for lack of a better word I mean I feel bad calling something dumb but I guess it's not alive so it's okay but like I get that it would Plateau but why would it regress well you know so many of these systems are just based on fundamentally learning from what's openly available on the internet and let's be honest a lot of what's out there on the Internet isn't that great and if we run out of data to train because you've looked at the entire internet and all you can do then is look at the data that has been generated on your own system you're you're in essence uh drinking from the same well that you're uh that you're uh you know feeding into and so you know it's not surprising at all to me that you know I think in one report you know one of the llms you know could do math very well a few months ago and now all of a sudden it can't even do two plus two reliably and that's you know that speaks to this fact that the data that they're training these on is is not growing at the rate that the systems themselves are growing so uh Sultan there's been kind of there had been some criticism about where Apple was with regard to AI especially when you think about the context of chat GPT and Microsoft being a front runner now there and now Apple apparently creating its own According to some reporting Apple GPT or what their Ajax I think that's what the name is uh that they're calling what they're working on uh who do you think is going to be the winner in this space well I hope you're not asking me to bet against Apple because that seems like a dangerous thing to do but you know apple has a long history it's part of their culture of an organization you know they weren't the first personal computer IBM was they weren't the first smartphone you know we all had blackberries before that I guess I'm not surprised at all that they've taken a slightly slower approach a more engineering Centric approach than many of these other organizations but they also as one of the few trillion dollar tech companies out there have a massive consumer base and the minute they decide to release something like that Millions if not hundreds of millions of people will be using it within 24 hours and you know when you have that kind of scale you have to be very thoughtful about introducing a new feature that that has the support on the back end you know one thing that most people might realize is is our growth in artificial intelligence is limited by the numbers of CPUs and gpus that can actually do the the math that makes them work and if all of a sudden you have to add 100 million users to an AI system you have to have a fair amount of infrastructure behind that and it's possible possible that they just don't have it yeah well that explains the enthusiasm and the demand for NVIDIA chips and shares for example but again to sort of talk about where we are in the cycle with apple sort of being a little more cautious about it maybe we should also take a lesson from Apple I mean I think about all of the hype cycles that we have gone through in the past decade from crypto to the metaverse even to full self-driving where the talk was big and the yield either didn't materialize at all or has been much much slower than the proponents would have said and I kept hearing at the beginning of this year Well AI is different than that that you know the the it's such a big thing and can affect so much was that just fundamentally wrong or is it just the timeline is way more extended well you know that's an absolutely fantastic way to frame it and I would say it's the timeline you know the metaphor that I use is the cloud you know the phrase the cloud was really coined in 2007-2008 and it's taken many organizations you know through 2023 to to even have a cloud strategy and so you know the US government where I work you know we weren't fully on the cloud by a long shot um still aren't and there are time considerations that most people don't realize when you're talking about impacts of massively scaled Technologies so if it if you go from 2008 to 2023 and we aren't still fully on the cloud with AI being in my opinion an even more impactful technology I wouldn't be at all surprised if we're talking about a multi-decade journey to really realize the benefits of artificial intelligence and Sultan you talked about the data to train it earlier and how you know I guess if you put stupid in you get stupid out where does the responsibility lie with regard to trade training AI especially when you think about the potential or the use case within law enforcement and just the legal field in general to prevent against discriminatory bias I mean discrimination not just in law enforcement but also in basic things like you know making credit or lending decisions you know there is a tremendous risk attached to that and so you know the Biden Administration you know approaching artificial intelligence as a risk which is what they've done you know shows that they have a they have thought about this front and center the challenge is is just like the internet itself government can't do it alone the private sector can't do it alone Academia can't do it alone all of those organizations including you know the average American the average human need to be part of this discussion because the current laws the current regulations don't take these kinds of Technologies into account and we need new systems new regulations new ways of thinking about this in my opinion and we need to build that and that for the most part doesn't exist yet and so you know hearing that some companies will you know uh put some watermarking in there or choose to collaborate in a in a non-official capacity with the US government is a great step but that doesn't actually solve a broader problem

Channel: Yahoo Finance

Views: 6,185

Keywords: Yahoo Finance, Personal Finance, Money, Investing, Business, Savings, Investment, Stocks, Bonds, FX, Currencies, NYSE, Equities, News, Politics, Market, Markets, Yahoo FInance Premium, Stock market, AI

Id: 8gUe2wohfgc

Length: 8min 14sec (494 seconds)

Published: Sun Jul 23 2023