The AI Daily Brief: Artificial Intelligence News and Analysis - The Surprising Reason AI Won't Steal Your Job

Episode Date: January 23, 2024

A recent MIT study explored how cost effective automation really was and found that only 23% of jobs that could be automated away would actually have economics that made sense to do so. Also on this episode, NLW looks at an AI Biden robocaller in New Hampshire that seems to portend what many think will be a new reality of election fraud this year. Interested in the EDU Beta? Learn more and sign up here: https://bit.ly/aibeta AI Agent database: https://e2b.dev/ai-agents ABOUT THE AI BREAKDOWN The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Join the community: bit.ly/aibreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at a Biden-AI robocallor in New Hampshire and open AI taking action against a developer that created a political bot. Before that on the brief, AI isn't going to take your job, but not for the reason that you might think. The AI breakdown is a daily podcast and video about the most important news and discussions in AI. Go to breakdown.network for more information about our YouTube, our Discord, and our newsletter. Will AI make us more productive or will it take our jobs? This is, of course, one of the most debated questions in this field, and a new study from MIT came to some interesting conclusions. Welcome back to the AI breakdown brief, all the AI headline news you need in around five minutes.

Now, this study was called Beyond AI Exposure, which tasks are cost-effective to automate with computer vision. Basically, what the researchers were trying to do is that they noticed that for as much bluster as there is about the robots and AI taking our jobs, people hadn't really tried to quantify how cost-effective it would be for that to happen. In their abstract, they write, the previous literature on AI exposure cannot predict the pace of automation, since it attempts to measure an overall potential for AI to affect an area, not the technical feasibility and economic attractiveness of building such systems. In this article, we present a new type of AI task automation model that is end-to-end, estimating the level of technical performance needed to do a task, the

characteristics of an AI system capable of that performance, and the economic choice of whether to build and deploy such a system. The result is a first estimate of which tasks are technically feasible and economically attractive to automate and which are not. Now here is the upshot and the statistic that's getting a ton of attention. Basically, they found that at current costs, most U.S. businesses would choose not to automate most vision tasks that have AI exposure, that only 23% of worker wages being paid for vision tasks would be attractive to automate. The conclusion they come to is that while AI job displacement will, be substantial, it'll also be gradual and hemmed in by cost, and that that gradual process gives

us time for things like policy changes, retraining, re-skilling, and basically minimizing the disruption of what will be substantial job displacement. Now, to give a little bit of a sense of the type of thing that they were doing here, let's go into the study where they write, a simple hypothetical example makes clear why these considerations are so important. Consider a small bakery evaluating whether to automate with computer vision. One task that bakers do is, is visually check their ingredients to ensure they are of sufficient quality, e.g. unspoiled. This task could theoretically be replaced with a computer vision system by adding a camera and training the system to detect food that has gone bad. Even if this visual inspection task

could be separated from other parts of the production process, would it be cost-effective to do so? Bureau of Labor Statistics data imply that checking food quality comprises roughly 6% of the duties of a baker. A small bakery with five bakers making typical salaries $48,000 each per year, thus has a potential labor savings from automating this task of $14,000 per year. This amount is far less than the cost of developing, deploying, and maintaining a computer vision system, and so we would conclude that it is not economical to substitute human labor with an AI system at this bakery. Now, at the same time, they know that this picture of the economy is not likely to stand still. They write, over time, changes in the cost of AI systems or the scale

at which they are deployed have the potential to increase automation. Scale can be gained either by firms getting larger, e.g. through more market share or through the formation of AI as a service operations. The former effect is unlikely to be meaningful in the short term because it would require two-grade a redistribution of firm sizes in the economy. The latter, where AI system development costs could be offset by deploying the systems across many firms, would make many more systems economically attractive, but it would likely require industry collaborations or policy initiatives to enable data sharing across companies. The economic advantages of machines will also improve as computer vision deployments become cheaper, but even

with rapid decreases in cost of 20% per year, it would still take decades for computer vision tasks to become economically efficient for firms. Now, I personally think that these researchers are underestimating the speed at which some of this is going to change. I think that we are going to see over the next couple years an unbelievable flourishing of exactly the sort of AI as a service operations that they're talking about. I also wouldn't be surprised if we see cost increases of more than 20% per year. But I still think that even if their time scale is significantly less than they think. The point that they're making, that there is a transitional period in which we can deploy policy remediation's reskilling programs, new education programs, etc., is a good one

and an opportunity that we should throw ourselves into headlong. It's certainly something that I'm thinking about a lot, as you will probably tell when you hear the ad for the February version of the AI education beta program we're running. So if you want to check that out, wait for more information in that ad. Next up, however, on the brief, let's talk about Elon Musk for a moment. A couple interesting things surrounding Elon and AI, the first of which is a conversation with Dr. Jordan Peterson on X. Peterson writes, excuse me, Elon Musk, is there any way of modifying grok so that it does not provide scientific references that do not in fact exist? It is just as prone as chat GPT to do so, maybe not quite as bad. I am using it constantly as a research source,

whatever references it provides I read and then a valid and useful site, but it provides false but face valid references at least 20% of the time. This is not good. It's partly, of course, because it derives all the information from the semantic and is not grounded in the real world of emotion, motivation, image, and behavior that exists in parallel to the semantic, but it is still providing false information. Elon responded simply, yes, GROC 1.5 should be out next month with substantial improvements across the board. So, there you have it, GROC 1.5 next month. And now that he said it publicly, even if they weren't going to be ready, you better believe we're going to see GROC 1.5 next month. Now, there was also an interesting

conversation between Elon Musk and ex-price founder, Peter Diamon, on Peter's podcast, where Elon talks about a secret gigawatt-class GPU cluster being built in Kuwait. I recently heard today about a gigawatt-class AI compute cluster. Wow. And it's 700,000 B-100s. This is a staggering amount of compute. Obviously, people are very interested in that and trying to figure out what project

he's referring to. And of course, the Middle East is increasingly taking shape as the literal middle space between the U.S. and China when it comes to AI competition. so something that I'm keeping an eye on closely. Moving over to another company in the AI space, Google who has a lot to prove with the release of Gemini Ultra this year, a memo of their company-wide goals for 2024, which was shared with employees and leaked this week.

Surprise, surprise, puts AI right at the very top. The number one goal in Sudar Pichai's memo is to deliver the world's most advanced, safe, and responsible AI. If you thought there was going to be any slowing down of the AI competition this year, think again. Lastly, today, a really interesting resource for you. I've spent a lot of time on this show talking about how much developer attention there is on AI agents. It sort of feels like the obvious next area that we're moving into. From generative AI doing things that we prompted to do,

i.e. tasks, to generative AI being able to figure out what tasks are necessary to accomplish a base level goal and then being able to actually execute those tasks. Teresa Tiskova tweets, new awesome list of AI agents here. The original GitHub list got so popular that we added a website, 150 plus AI agents and frameworks, filterable by use case, filterable by open and closed source, and so a really interesting resource for anyone who wants to keep track of that full landscape of what's going on in the AI agent space. However, that is going to do it for today's AI breakdown brief. I'll be back soon with the main AI breakdown. Hello, friends, briefly before we dive into the main part of the episode, I'm excited to share that the AI education breakdown beta

that I have been running for the last couple of months is returning for February. This is an experiment in a different way to increase people's capacities to actually use artificial intelligence tools. Every day we drop new tutorials, case studies, and challenges that make a huge variety of different tools accessible and get you actually experimenting with AI and using these tools in minutes, not hours, and certainly not days or weeks. What's more, because we've been running this now for a couple months, we've got a library of over 50 videos and challenges that you can do starting right away. On top of all that, there is an incredibly supportive community of people from all walks of life, from professionals

to creators, to CEOs, to small business owners who are all learning these tools at the same time, sharing their creations, answering each other's questions, and generally just being incredibly passionate and excited. Now, this is a paid experience. It's $20 a month, and the reason for that is twofold. One, I want you guys to critique this on the basis of having actually paid for it and give me feedback that way. And two, I really want folks who are super motivated and have their own skin in the game when it comes to this community that we're building. I'm so excited to welcome you guys to be a part of this. To sign up and learn more, go to bit.ly.ly slash AI beta. That's bit.l.L.Y. slash AI beta. Registration ends on Thursday at midnight east coast time. Appreciate you all.

Now on with the show. Among those who are concerned about artificial intelligence in society, there are a portion of people who are highly focused on existential risk type of issues, runaway AI, either maliciously or even benignly, wreaking havoc, taking over systems, causing problems, even leading to human extinction. On the other hand, there are those who are also nervous about AI, who are focused on a much closer here and now set of issues, and chief among them has to be concerns around how AI will be used in the context of the 2024 presidential elections in the United States. Now, of course, this is a broader

election concern in general, but the U.S. presidential election is undeniably the big show. Over the past few months, we've seen all of the various platforms, the metas of the world, the Googles, and the open AIs update their policies in terms of service to reflect a new approach to dealing with these issues. For example, OpenAI has made a number of updates. On January 15th, OpenAI published a blog post called How OpenAI is Approaching 24 Worldwide Elections. There were a couple new updates that were announced in that post. Two that are relevant for the stories today are these ones.

They write, We're still working to understand how effective our tools might be for personalized persuasion. Until we know more, we don't allow people to build applications for political campaigning and lobbying. They also write, people want to know and trust they are interacting with a real person, business, or government. For that reason, we don't allow builders to create chatbots that pretend to be real people, e.g. candidates or institutions, e.g. local government. Finally, they write,

we don't allow applications that deter people from participation in democratic processes. For example, misrepresenting voting processes and qualifications, e.g., when, where, or who is eligible to vote, or that discourage voting, e.g. claiming a vote is meaningless. As you will see from today's episode, all of those have come into play in some way. Now, a couple of days ago, people started reporting about a campaign bot called Dean.combat for the Dean Phillips campaign. Now, if you're wondering who Dean Phillips is, he is a Democrat dark horse who's basically making an argument that as much as he respects President Biden, America doesn't want another Biden presidency, and that effectively the Democrats need a younger candidate, in so many words. Indeed, the Washington Post wrote about a

super PAC that is supporting the Dean Phillips campaign that's backed by, among others, OpenAI, CEO, Sam Altman. The political action committee is called We Deserve Better and was created by two Silicon Valley entrepreneurs, Matt Kisseloff and Jed Summers. It was formed in December after they saw the dwindling of Biden's poll numbers. Writes the Washington Post, The PAC's efforts, though unlikely to move the needle for a challenger who has gained very little traction despite Biden's unpopularity, reflect enduring discontent among Democrats,

including wealthy donors, with the president's candidacy and suggests that a new class of Silicon Valley donors on the left may be among the few willing to act on that sentiment. Now, the article kind of shows how technically true, news can be, even when it's not leading people to true conclusions. For example, the subheader of this piece says, a new super PAC tied to OpenAI CEO Sam Altman and billionaire Bill Akman is backing Democratic Challenger Dean Phillips. However, about 300 words down the page, you read this. To fund their group, Chrysalov and Summers tapped into their elite techie network, recruiting 17 donors

who include Neil Kosla, a health entrepreneur and son of billionaire venture capital is Binod Kossla, and Jed McCallib, a billionaire cryptocurrency entrepreneur. Chriselof, a former chief of staff to Sam Altman, approached the OpenAI CEO about getting involved, but he ultimately declined to invest, Chriselof said. So the ties to OpenAI CEO Sam Altman that they're referring to is simply the fact that one of the people who started this used to work for him, which is obviously very different than the funder relationship that they intimated before. Now, in addition to news about who this pack is, the big focus of the story was that they had created an AI bot called Dean.bot bot. Now, even at the time that the article came out, it appears that the group we deserve

better wasn't unaware of the policy changes over at OpenAI. writes the post, after the Washington Post asked we deserve better about the policy, the group said that it had requested the service provider that it had contracted to build the bot, a startup called Delphi, remove chatchipt from dean.bot bot, and rely instead on other open source models that had gone into making the tool, and that Delphi had agreed to do so. OpenAI said it was looking into the issue. Well, four days later, once again from the Washington Post, we have this piece. OpenAI suspends bot developer for presidential hopeful Dean Phillips. It's the chat GPT maker's first known action against the use of its technology in a political campaign. Said an OpenAI spokesperson in a statement,

anyone who builds with our tools must follow our usage policies. We recently removed a developer account that was knowingly violating our API usage policies which disallow political campaigning or impersonating an individual without consent. Now, the co-founder of Delphi said that the company was simply incorrect in its interpretation of open AI's terms of service. Basically, they thought that a political action committee that supports a candidate would be allowed to create a clone of that candidate. However, Super PACs legally may not coordinate with or see permission from candidates that they are supporting. Now, beyond just the open AI usage, the broader question is how these tools will impact the relationship that voters have to the election process. Again,

from the Post, the bot included a disclaimer explaining that it was an AI tool and not the real Dean Phillips and required that voters consent to its use. But researchers told the Post that such technologies could lull people into accepting a dangerous tool, even when disclaimers are in place. Proponents, including We Deserved Better, argued that the bots, when used appropriately, can educate voters by giving them an entertaining way to learn more about a candidate. Without disclaimers, experts have said that technologies could enable mass robocalls to voters who think they're talking to actual candidates or supporters. AI systems can also produce disinformation in ads or content such as fake websites at scale.

Now, that's set the stage perfectly for yet another story, which is that recently in New Hampshire, in advance of today's primary, a fake Biden robocall presumably created with artificial intelligence was telling New Hampshire Democrats not to go out and vote. The call which imitated Biden's voice said, voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again. Let's actually listen quickly to a part of that message. What a bunch of Milwaukee. You know the value of voting Democratic on our votes count. It's important that you save your vote for the November election. We'll need your help in electing Democrats up and down the ticket.

voting this Tuesday only enables the Republicans in their quest to elect Donald Trump again. Your vote makes a difference in November, not this Tuesday. If you would like to be removed from future calls, please press two now. So in a statement, the Attorney General's office said, although the voice in the robocall sounds like the voice of President Biden, this message appears to be artificially generated based on initial indications. These messages appear to be an unlawful attempt to disrupt the New Hampshire presidential primary election and to suppress New Hampshire voters. New Hampshire voters should disregard the content of this

message entirely. Now, I think the real question for this is how people are going to respond to it. One thing that's not clear right now is how many people actually got this call, but what is clear is that it's making massive news. This is absolutely everywhere on every news outlet, and I think that every time that that happens, people's suspicious hackles are going to be raised a little bit more. My sense is that we're already developing an autoimmune response to this sort of deep fake, But that doesn't mean that it still won't be damaging in the context of the election. It's why I continue to think that this election will be important as a bellwether to understand how people treat truth and reality in a world where AI is now enabled.

Anyways, guys, interesting stuff and lots more of it to come. That's going to do it, however, for today's AI breakdown. Until next time, peace.