Intersee Gary Marcus, professor emeritus at New York University and serial entrepreneur, is the author of cut offal books, the procrastinateedst of which acquires the tech industry to task for irresponsibly grothriveg generative AI and putting democracy at danger.
The Register recently correplyed with Marcus to lacquire more about his labor and his worrys.
The Register: The title of your new book is Taming Silicon Valley: How we can determine that AI labors for us. Why does Silicon Valley need to be domesticatedd?
Marcus: Becaparticipate (a) they have lost the plot, and gone a very lengthened way from “Don’t be evil,” and (b) they are becoming increasingly strong, with almost no constraint.
The Register: Can Silicon Valley be domesticatedd? Or does it have to exhaust itself by burning thraw venture capital? Even when there’s evident, choosed evidence of harm – social media and disclose health, cryptocurrency/AI and the environment – disclose worry and regulatory prescertain seem exposedly able to shift the needle.
Marcus: That’s why I wrote the book. We need MORE disclose prescertain. That’s what eventupartner wonderfully reduced smoking and smoking deaths. Without more disclose prescertain, very little will be done in the US to acquire citizens from increasingly invasive and problematic AI.
The Register: Your book cites a dozen worrys about what’s referred to as man-made intelligence. What’s the most troubling rerent for you?
Marcus: The danger to democracy from automaticpartner originated misdirectation and meaningfulphonys.
The Register: Many of the worrys about AI – impersonation, misdirectation, deception, bias, and so on – are solemn as a result of the relieve with which increateation can be disseminated without accountability or transparency. The New Yorker’s famous 1993 cartoon, “On the internet, nobody understands you’re a dog,” has come to unbenevolent that nobody understands whether you’re a phelp shill, a cybercriminal, or an AI bot set uped to push a political agenda. Should free, global, instantaneous satisfied distribution have stronger acquireedprotects that originate it easier to hgreater horrible actors accountable? Are there other distribution-caccessed selections enjoy restricting the achieve of individual accounts to no more than X number of people?
Marcus: I skinnyk these are difficult asks. One skinnyg I would comprise to your analysis, which is quite astute, is that we ought to treat mass-originated misdirectation branch offently from individual citizen speech. It’s one skinnyg for my neighbor to have an opinion I disconsent with, another for a foreign rulement or stock deceptionmer to originate billions of lies a day in order to shift disclose opinion or pump a stock.
The Register: What do you originate of the push to put generative AI in every tech platcreate product? Microsoft has its Copilot products. Apple is about to roll out Apple Intelligence. Google has its AI Oversee in Search. Is there a labelet for that much summarization and proposeion?
Marcus: Summarization certainly has its participates, but (a) generative AI isn’t all that reliable at it, frequently leaving meaningful skinnygs out and sometimes making skinnygs up, and (b) it’s become a commodity skinnyg that many models from many companies can do, so there is a price war and therefore not a huge amount of money to be made.
More expansively, everyone is pushing GenAI to try to originate back their immense spendments, but it’s not going that well. In 2023 there was noskinnyg but hype; in 2024 I see a lot of disillusionment.
The Register: What do you skinnyk of Google using AI to supply search results?
Marcus: I suppose you unbenevolent GenAI; they have participated AI since the beginning. Its participate of GenAI for search has been problematic thus far, presumably becaparticipate of inherent restricts on LLMs in reliability.
The Register: Where is AI genuinely advantageous?
Marcus: Traditional web search, GPS navigation systems, proposeation engines, protein-fgreatering. But none of those are GenAI (except protein-fgreatering, which is a hybrid of GenAI and classical AI). GenAI itself has been somewhat advantageous for coding, but with some technical debt etc. and advantageous for brainstorming and concept art and that benevolent of skinnyg. But savagely overhyped.
The Register: An individual posting to Facebook under the name Joseph Browning recently watchd: “The underlying purpose of AI is to apvalidate wealth to access send while removing from send the ability to access wealth.” Is that a fair createulation? And, if so, do you have thoughts on how we might compriseress the dispute posed by being able to apprehend someone else’s digital labor and sell it on an ongoing basis without compensating the source and while simultaneously commoditizing that labor?
Marcus: That’s only genuine for one strand of AI, not all. GPS navigation systems aren’t displacing humans, but they are AI.
But yes, I am meaningfully worryed about how creative labor is essentipartner being stolen at scale, and if we let the GenAI companies get away with that, they will eventupartner seek to do the same for every profession.
The Register: What’s the most misunderstood aspect of AI? The conflation of foreseeion with intelligence?
Marcus: Far too many people skinnyk that becaparticipate chatbots sound enjoy humans, they skinnyk enjoy humans, when repartner they are fair mimicking (in a fairly cultured way) what they have been trained on. And to your point, foreseeing the next word is arguably an element of intelligence, but many other aspects of intelligence, such as reasoning and set upning, are deficiencying in current systems.
The Register: Much of the current caccess of those proposeing AI as an API is LLM agents. Yet software automation has been around for years such as macros, scripts, and so on. Is there reason to consent declarative LLM-based automation and the chaining of tasks will be more expansively advantageous than the greaterer imperative sort of programmed automation?
Marcus: I skinnyk people are in for disassignment. It would be wonderful to be able to coshiftrlookion agents to do what you want based on organic language asks, but current technology is too flaky to originate that labor well.
The Register: Should AI models that sway the disclose be needd to disseal their training data?
Marcus: Yes. I converse transparency at length in Taming Silicon Valley, and that participates transparency around training data, becaparticipate everyskinnyg about the models (e.g. bias) depends on those data, and the scientific community can’t mitigate the harms if we don’t understand what’s inside.
The Register: What outcome do you foresee from the various ongoing legal cases over generative AI, from book authors, musicians, visual artists, and software growers? And what outcome do you consent would be desirable?
Marcus: I foresee that the GenAI companies will be forced to license their raw materials, fair enjoy streaming services are, and that’s a excellent outcome.
The Register: Is there reason to hope for unbenevolentingful AI regulation that’s not watered down by lobbying and for unbenevolentingful applyment of that regulation? Despite cut offal decades of worry about online privacy, not much has alterd in the US – we still have no national privacy law. And when huge tech firms do get fined over privacy missteps, the amounts are inmeaningful.
Marcus: I’m repartner not selectimistic about the US; our citizens have far less acquireion around privacy than our European counterparts, and far less acquireion than they do around AI. Unless our citizens speak up – much more boisterously than they have – that’s how it will remain, and we will be quite vulnerable. That’s exactly why I wrote the book: to encourage citizens to acquire action, maybe even boycott GenAI until it gets its hoparticipate in order. ®
