Why Google Is Intentionally Limiting the Power of Bard AI

0:00Today on the AI Breakdown, what Google's recent public conversations about AI suggests about the state of the field. Before that on the brief, the White House hosts its first classified AI briefing. 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. Welcome back to the AI Breakdown Brief, all the AI headline news you need in five-ish minutes or less. Today is a highly geopolitical, geostrategic day in AI, and we kick off with the big news that senators will receive their first ever classified AI briefing from the White House later today.

0:37Now, if you've been following along, you'll know that AI is a growing issue when it comes to the priorities of the White House, Senate, and Congress. In particular, Senate Majority Leader Chuck Schumer has made it his goal to lead a number of different AI-related efforts, including recently a dramatic speech held at the end of January. In that speech, he said, if applied correctly, AI promises to transform life on earth for the better. It will reshape how we fight disease, tackle hunger, manage our lives, enrich our minds, and ensure peace. But there are real dangers that present themselves as well.

1:06Job displacement, misinformation, a new age of weaponry, and the risk of being unable to manage this new technology altogether. That same week, President Joe Biden also dramatized the rise of AI, saying we'll see more technological change in the next 10 years than we saw in the last 50 years. Now, at the same time, Schumer said that he was working on what he called very ambitious legislation, and part of the path to get there was a set of briefings for senators that would happen over the rest of 2023. The briefing that they have today is one of those promised sessions. Now, the reason that this briefing in particular is classified is that they're going to be dealing with issues of national security, how AI relates to geopolitical contests, how the White House is using AI, and more sensitive topics like that.

1:44Senators will be briefed by the Director of National Intelligence, the Deputy Secretary of Defense, the Director of the White House Office of Science and Technology Policy, the Director of the National Geospatial Intelligence Agency, and the Defense Department's Chief Digital Officer. This is a bipartisan briefing being hosted by two Republicans and another Democrat, in addition to Chuck Schumer. Now, this briefing on the national security implications of AI comes at an interesting time. The discourse is increasing pretty significantly about the national security implications of AI, as well as the military implications of AI.

2:16On Monday, Netflix premiered a new documentary called Unknown Killer Robots, and as this Guardian review puts it, quote, The future of AI will fill you with unholy terror. Now the goal of this documentary is to try to show what the new technologies are, while also questioning whether we're getting into a dangerous, literal AI arms race. I think the Guardian Review was insightful when they wrote, The dilemma surrounding almost all military inventions, perhaps almost all inventions full stop, is what is slightly grandly called the dual-use problem. On the one hand, you've got drones and robots who can clear buildings without risking soldiers' lives.

2:48On the other, you can weaponize them, autonomize them, and use them to take out entire villages without anyone getting their hands dirty. What is the sense of detachment likely to do to the level of carnage in a war overall? Former U.S. Defense Secretary Bob Work doesn't think, quote, human intervention and kill decisions will ever change. I cannot help but pause for a moment to suggest respectfully that either the good colonel has never met humanity, or that he is the program's equivalent of the flight attendant urging people to calm down as the passenger jet plummets to its fiery doom. Netflix, however, isn't the only place where this discussion around AI and military applications is showing up.

3:21Time ran a large piece recently about how tech company Palantir is helping change the face of warfare. The piece, called Literally How Palantir is Shaping the Future of Warfare, writes, Alex Karp, Palantir's CEO, has argued that, quote, The piece goes on. One idea I heard is that warfare may increasingly take place as a complex simulation within algorithmic systems. The process may have some deterrence powers. Two opponents might reach the same conclusion about the outcome, preempting any need to trigger a conflict in the physical world. Is this utopian? Probably. The most likely scenario is an algorithmic arms race happening at superhuman speed.

4:02Here China, rather than Russia, is the real opponent. Taiwan, rather than Ukraine, is where the algorithm takes over. As you heard just a moment ago, when it comes to AI and the military, one of the U.S.'s chief concerns is China. In mid-May, Bloomberg published a piece, China's military use of AI raises alarm for Congress. The piece begins, China's embrace of artificial intelligence for warfare has touched off alarm bells everywhere from Silicon Valley to the Pentagon. Former Google chief executive officer Eric Schmidt is among those raising concerns, and he testified at a House hearing about China Wednesday evening as head of an initiative that's focused on speeding the U.S.

4:37defense establishment's adoption of AI. The piece goes on to profile a report from Schmidt's organization, which is called the Special Competitive Studies Project, and as Bloomberg puts it, a report from that organization describes a 30-year effort by China to study U.S. combat operations with the intention of being able to puncture its military might now with the help of AI. Schmidt basically argued at this hearing that even if the U.S. is currently ahead of China when it comes to AI, it seemed likely to him that they were dedicating much more resources to advancement in the strategic and defensive use of AI.

5:07Now, we haven't really gotten much more information about the military use of AI since then, but the Financial Times did report yesterday about some new efforts in China to implement AI rules that would effectively amount to a licensing regime for generative AI companies. The Financial Times writes, The Cyberspace Administration of China, the powerful internet watchdog, aims to create a system to force companies to obtain a license before they release generative AI models, said two people close to Chinese regulators. The requirement tightens draft regulations issued in April, which said groups would have 10 working days to register a product with authorities after launch.

5:39Now, people often use China as the boogeyman when it comes to AI safety. The standard argument goes, well, we can't pause or stop or slow down our development of AI because if we do, China won't. Still, this Financial Times piece points out that there is an inherent tension in AI. As they put it, Beijing is struggling to reconcile an ambition to develop world-beating technologies with its long-standing censorship regime. Matt Sheehan, a fellow at the Carnegie Endowment for International Peace, said, it's the first time that authorities in China find themselves having to do a trade-off between two Communist Party goals of sustaining AI leadership and controlling information.

6:13The FT goes on, one person close to the CAC's deliberation said, if Beijing intends to completely control and censor the information created by AI, they will require all companies to obtain prior approval from the authorities. But the regulation must avoid stifling domestic companies in the tech race, so authorities are wavering. I think the decisions that the Chinese government makes when it comes to this particular issue, how much to lean on the side of global competitiveness of their domestic AI versus command and controlling the outputs of that AI, is going to have a huge deterministic impact on whatever AI arms race there actually is.

6:46Meanwhile, over in the US, we have reports that Google is intentionally limiting BARD's capabilities, And we're going to get into that in the main AI breakdown, which is coming up soon. However, that's going to do it for today's brief. If you're enjoying it, hit that like button, or even better, hit the notification button so you don't miss an episode. And I'll be back soon with the main AI breakdown. Do you think society is prepared for what's coming? You know, there are two ways I think about it. On one hand, I feel no, because, you know, the pace at which we can think and adapt as societal institutions compared to the pace at which the technology is evolving, there seems to be a mismatch.

7:26On the other hand, compared to any other technology, I've seen more people worried about it earlier in its life cycle. So I feel optimistic the number of people who have started worrying about the implications, and hence the conversations are starting in a serious way as well. The clip you just heard was part of a recent 60 Minutes interview with Google CEO Sundar Pichai. The interview was a part of a set of recent communications from top leadership at Google that help us understand how the company is thinking about AI in general. Given that they are one of the small number of companies alongside OpenAI, Microsoft, Meta, and not very many others who are part of the intensive corporate AI battles that led to, for example, the recent defection of Jeffrey Hinton from Google, with concern around what those battles would do for the future of AI safety and AI risk, understanding how Google is thinking about AI issues and what they want to broadcast publicly, I think is pretty relevant.

8:27Now, when it comes to that 60 Minutes interview, one of the aspects of it that most people have picked up is the idea expressed in this Fox Business headline that Pichai said that Google is intentionally limiting BARD AI's public capabilities. The context for that part of the conversation came when Pichai said, it would be possible with AI to create, you know, a video easily, where it could be Scott saying something or me saying something and we never said that. And it could look accurate. But you know, on a societal scale, it can cause a lot of harm. So obviously here Sundar is talking about deepfakes.

8:57The interviewer, Scott Pelley, responded with a question saying, is BARD safe for society? To that, Sundar said, the way we have launched it today as an experiment in a limited way, I think so. But we all have to be responsible in each step along the way. Pelly asked for clarification, saying, you are letting this out slowly so that society can get used to it. Sundar said, that's one part of it. One part is also so that we get the user feedback and we can develop more robust safety layers before we deploy more capable models. In that same interview, Sundar really put the stakes of AI pretty clearly.

9:26He said, I've always thought of AI as the most profound technology humanity is working on, more profound than fire or electricity or anything that we've done in the past. As part of the conversation, Sundar also described the need for regulations. He said, you're going to need laws. There have to be consequences for creating deepfake videos which cause harm to society. Anybody who has worked with AI for a while realizes this is something so different and so deep that we would need societal regulations to think about how to adapt. Now, the conversation about deepfakes is definitely heating up right now.

9:54Sapien's author Yuval Noah Harari recently made headlines when he said that AI firms should face prison if they fail to guard against the creation of fake profiles on their platforms. Harari was addressing the UN's AI for Good summit last week when he said, It is possible for the first time in history to create fake people, billions of fake people. If this is allowed to happen, it will do to society what fake money threatened to do to the financial system. If you can't know who is a real human, trust will collapse. Maybe relationships will be able to manage somehow, but not democracy. Harari went on, What happens if you have a social media platform where millions of bots can create content that is in many ways superior to what humans can create?

10:31More convincing, more appealing. If we allow this to happen, then humans have completely lost control of the public conversation. Democracy will become completely unworkable. And whether you agree with these suggestions or not, give Harari credit for at least talking about what he thinks some of the answers might be. For example, he said that if tech executives face 20-year jail sentences for not finding ways to prevent these sort of fake profiles, he predicted they would, quote, quickly find ways to prevent the platforms from becoming overwhelmed with fake people. He also suggested that companies in the AI space should be legally required to commit 20 % of their investment spending to researching AI safety risks and how to address them.

11:04Now, for some, one of the counterweights to the power accruing to these big AI companies like Google and OpenAI is more robust open source models. In fact, a couple of months ago, an internal Google memo was leaked that claimed, quote, we have no moat and neither does OpenAI. That memo started, we've done a lot of looking over our shoulders at OpenAI. Who will cross the next milestone? What will the next move be? But the uncomfortable truth is we aren't positioned to win this arms race and neither is OpenAI. While we've been squabbling, a third faction has been quietly eating our lunch. I'm talking, of course, about open source.

11:37Plainly put, they are lapping us. Things we consider major open problems are solved and in people's hands today. The examples that this anonymous author pointed to include LLMs on a phone, scalable personal AI, responsible release, and multimodality. Now the genesis for a lot of this was the developments that happened in the wake of MetaLlama's model being fully leaked. The memo author goes on, while our models still hold a slight edge in terms of quality, the gap is closing astonishingly quickly. Open source models are faster, more customizable, more private, and pound for pound more capable.

12:06They are doing things with$100 and 13 billion parameters that we struggled with at$10 million and 540 billion parameters. And they are doing so in weeks, not months. The implications, he said, are that they have no secret sauce, that people will not pay for a restricted model when free unrestricted alternatives are comparable in quality, and one of their biggest arguments, that giant models are slowing Google down. Now, this memo got a ton of attention in the AI space when it was released, and it appears that it caught the attention of Google's internal teams as well. In a recent interview with Decoder, Demis Hassabis, the CEO of Google's DeepMind, said, I think that memo was real.

12:41I think engineers at Google often write various documents, and sometimes they get leaked and go viral. I think it's interesting to listen to them and then you've got to chart your own course. And I haven't read that specific memo in detail, but I disagree with the conclusions from that. One of the interesting parts of the discussion had to do with how and why ChatGPT was the kickoff moment that led to this AI explosion. The interviewer says it feels like the ChatGPT moment was really rooted in the AI being able to do something that regular people could do. There's this turn where the computer is starting to do things I can do and they're not even necessarily the most complicated tasks.

13:13Read this webpage and deliver a summary of it to me. But that's the thing that unlocked everyone's brain. And I'm wondering why you think the industry didn't see that turn coming, because we've been very focused on these very difficult things that people couldn't do, and it seems like what got everyone is when the computer started doing things people do all the time. Hassabis said, I think that analysis is correct. I think that is why the large language models have really entered the public consciousness because it's something the average person, that the Joe public, can actually understand and interact with.

13:39And of course, language is core to human intelligence in our everyday lives. I think that does explain why chatbots specifically have gone viral in the way they have. Now, Hassabis goes on to point out that other applications of AI like their AlphaFold might have had even more beneficial impact on the world so far, given in that case that it's used by all the big pharma companies to advance drug discovery, but that that's invisible compared to the average person on the street, who, as he puts it, doesn't know what proteins even are, and doesn't know what the importance of those things are for things like drug discovery.

14:07Now, another thing that Hassabis says is that although this is the thing that has captured people's imaginations right now, generative AI is just to use his term scratching the surface. Quote, There are a lot more types of AI than generative AI. Generative AI is now the in thing, but I think that planning and deep reinforcement learning and problem solving and reasoning, those kinds of capabilities are going to come back in the next wave after this, along with the current capabilities of the current systems. So I think in a year or two's time, if we were to talk again, we're going to be talking about entirely new types of products and experiences and services with never-before-seen capabilities.

14:37Now when it comes to what's next, Hassabis talks a lot about how the interfaces that are being created now, these natural language interfaces, are going to be the way that we interact with more specialized AIs that live underneath. He said, There's a whole branch of research going into what's called tool use. This is the idea that these large language models are large multimodal models. They're expert at language, of course, and maybe a few other capabilities like math and possibly coding. But when you ask them to do something specialized like fold a protein or play a game of chess or something like this, then actually what they end up doing is calling a tool, which could be another AI system that then provides the solution or the answer to that particular problem.

15:13And then that's transmitted back to the user via language or pictorially through the central large language model system. so it may be actually invisible to the user because to the user, it just looks like one big AI system that has many capabilities, but under the hood, it could be that actually the AI system is broken down into smaller ones that have specializations. I actually think that probably is going to be the next era. The next generation of systems will use those kind of capabilities. And then you can think of the central system as almost a switch statement that you effectively prompt with language, and it roots your query or your question or whatever it is you're asking to the right tool to solve that question for you or provide the solution for you, and then transmit that back in a very understandable way, again, using the best interface, really, of natural language.

15:50The interviewer then said, does that process get you closer to AGI? To which Hassabis said, I think that is on the critical path to AGI. The interviewer went on to ask how long Hassabis thought it would take to get to AGI, to which the CEO said, I think there's a lot of uncertainty around how many more breakthroughs are required to get to AGI, big, big breakthroughs, innovative breakthroughs, versus just scaling up existing solutions. And I think it very much depends on that in terms of timeframe. Right now, I would not be surprised if we approach something like AGI or AGI-like in the next decade.

16:18Now finally, returning to this open source question where we started, Hassabis says that while obviously they support open source, there are real questions as well. He said also safety questions about access to these very powerful systems. What if bad actors can access it, who maybe aren't that technical so they couldn't have built it themselves, but they can certainly reconfigure a system that is out there? What do you do about those things? I think it's all been quite theoretical till now, But I think that it is really important from here all the way to AGI as these systems become more general, more sophisticated, more powerful.

16:46That question is going to be very important about how does one stop bad actors just using these systems for things they weren't intended for, but for malicious purposes. Now, the last section of this interview that I'll note before I close has to do with the letter that Hassabis, along with so many others, signed from the Center for AI Safety. That was the one that basically said we need to treat the risks of AI alongside things like pandemic and nuclear proliferation. The interviewer said, basically, you've signed that letter and yet you are pushing on. Google's in the market. You've got to win.

17:13There's a tension there. Needing to win in the market with products and, oh boy, please regulate us because raw capitalism will drive us off the cliff with AI if we don't stop it in some way. How do you balance that risk? Now, I'll let you read the answer for yourself. But basically, Hassabis, one, acknowledges the tension, but two, makes it clear they have no idea what to do about it. In fact, the signing of this letter was basically just an acknowledgement that they don't know what's coming next, and that we need other actors who aren't the corporates who have corporate incentives to be involved in the conversation as some sort of counterweight.

17:43Anyways, it's a fascinating insight into how this one particular company thinks about AI, and given how important they are in the space overall, definitely worth your time. However, for now, that is going to do it for today's AI Breakdown. If you're enjoying this and you watched it on the YouTube, go subscribe to the podcast. And if you're listening to the podcast and you enjoyed it, go check out the YouTube channel. You can get links to all of that at breakdown.network. Appreciate you guys listening as always. And until next time, peace.