Anticipation, Resilience and Reliability: Three Ways AI Will Change DevSecOps … If We Do it Right | RSAC Virtual 2024

Thanks so much. Uh, thank you Mitch and Mark and Alan and the Textron team, especially Jody patient, Jody back there, uh, for their hospitality. Uh, I very much enjoyed listening to Zlatan and Amit, Matthew Knight.

I got, I scribbled down a couple of insights that I'll try to get to in my talk and Jason Collins from last night. Um, the whole notion of the prompt engineer as being a very important, um, uh, profession. I'm working with some prompt artists right now doing illustrations for, um, for some of my books, though I will not break the rice bowl of a human artist.

So the cover arts i, I pay for from a human artist. So you, that, those are the sorts of moral decisions we're having to make now if it's additive to what you're doing. But try not to break the rice bowl of pe uh, use AI to break the rice bowl of living people unless you're helping them to find a new way to earn their rice.

That's a little bit of an editorial there. Um, but in any event, uh, yeah, there's a lot to get through because this is an amazing age. I first attended RSA way long ago during the clipper chip era before most of you were born.

And, uh, I was telling people, calm down. This, this society will not permit anything like that silly idea. Um, I will say one thing though, I'm gonna mention a logic named Joe was a science fiction short story from the 19, late forties, early fifties.

There is an actual audio vis version that you should look up, A logic named Joe Maria Ster basically handled the, was presaged the prompt problem exactly in one of the characters asks this logic, um, how to murder his wife. And it's exactly what we went through during the great panic year of 2023. So look that up.

There's a lot of other highly, highly insightful, um, and, and prescient science fiction. Okay? I actually have changed the title of my talk, uh, based upon what I saw coming.

Um, and so I'm going to talk about how to do what Elon Musk and some others have been saying, teach your children well. And yet there's no basis for a mechanism how to teach your children. Well, in almost all of these discussions, you're finding again and again and again, uh, the word should, S-H-U-L-S-H-O-U-L-D, we should do this.

Um, your hackles should go up when you hear this word, because unless you come up with a method by which it could come true, it's not going to happen. Okay? So what are my qualifications?

Um, as mentioned, I've got the, you know, all this science fiction and all of that stuff, and there's one of my nonfiction books, vivid Tomorrow's about how science fiction is probably the reason why we're alive today. Um, and, uh, some of my novels have been called Highly Prescient. This novel Earth is on all the lists of the top 10 predictive novels, and this one just got reissued.

But I've also worked with NASA for the last 12 years in the, their Innovative and Advanced Concepts program funding, uh, very cutting edge, almost science, fictional, uh, research with, we have a very good track record, but the 19, the nineties clipper chip fuss at RSA and at Computers, freedom and Privacy led to my book, the Transparent Society Will Technology Force Us to Choose Between Privacy and Freedom. And it's still entirely, and there are no chapters in it that aren't, aren't still pertinent today. Alright, so this is a slide that I use in almost all my talks, so I'm gonna blaze through it because it's about the fundamental question of can our civilization survive?

It lists, um, six general categories of ways in which Toin Bee and Jared Diamond show that past civilizations, uh, failed to address their problems and therefore failed enemies in war, ecological suicide, suicidal meltdown. Does that sound fun? And the futile attract our state being outcompeted by others.

But there are new calamities that I often talk about, info ai, that's what we're talking about today, the Anthropocene Deadly Innovations and the Possibility of Renunciation. I'll be mentioning that too, because when people panic, they might turn away from the future. So this is all against the backdrop of two bookend concepts, which are the singularity and the Fermi paradox.

And I think a large enough majority of you out there know what these are, that I don't really have to explain them, but this, uh, talk will be available for you to follow up on. I should warn you that I treat PowerPoint as a writing vehicle. So if you can't see from back there, I recommend you move up because these, I violate every rule of PowerPoint, um, because I think that the it PowerPoint should be, you should be capable of reading it like a book.

Well enter, right? So a futurist that's me asks, what can both nature and human history teach us to help navigate this era of on regime change? Well, first off, we need to have context, and that's what I'm here for.

I step back sometimes like a text Avery character. I step back off a cliff and go about six or seven feet before I realize I'm out in the open. I step back and I talk about history, about the long time.

And so we need to think about how these disruptions have happened in the past, starting with the printing press. Gutenberg's, printing press made people in Europe realize that they were blind because they couldn't read. They didn't know it.

And so there it became a huge market for glass lenses. And some people poked with the glass lenses and invented telescopes and microscopes. So this, the printing press not only led to science and disruption that way by providing knowledge, but also by provoking a revolution in what we could see.

And since then, we have had revolution after revolution in what we can know, printing offloaded a large amount of our knowledge into our libraries, glass lenses, augmented what we can see and perspective what we can pay attention to. And so we had successive revolutions again and again and again that led to increases in what humans can know, see, and pay attention to. And always, there were two reactions.

There were Grouches who said, humans aren't built to be able to hand drink from this fire hose. This has gotta be mediated, it's gotta be controlled, or humans will go crazy. And there were always optimists who said, this is going to make us all bigger, better, more empathic.

We're going to know more and care more. And always the grouches were right. First, the printing press produced its primary product for about 50 years, were horrible religious tracks, slandering Protestants or slandering Catholics, and making the religious wars of Europe worse.

But eventually the optimists proved right. People learned more. They knew about the people over the mountain range.

They knew more and they became a little wider in their perspectives. Um, this has happened. It's created crises and quandaries and new concepts.

In 1930s, radios and loudspeakers almost killed us all because people weren't inoculated against the God-like amplification of guys like Mussolini and Hitler. And the only major difference in the English speaking world was that our gifted orators who hypnotized us, were on our side 1950s television. Um, it it, it continues in.

Now we're in the knowledge mesh, omni valence supervision. These are all doing exactly the same things. I'm not saying that we're going to succeed this time just because we succeeded the other times because those other times we barely succeeded and we had time.

There was time for the good stuff to overwhelm the bad stuff. Now we don't have time. We have to substitute for time.

We have to substitute sagacity. We have to actually think our way through this without taking the time for people to mature. And so what's next?

Ai. I give a lot of talks about human in human augmentation. It's an entirely different thread because this is some, an area in which we might achieve our soft landing by augmenting ourselves.

But in any event, there are all these logical, uh, methods to ai. I'm gonna list the eight that I as a well-informed. I do have a PhD.

I as a well-informed amateur in this field. This is my impression of the landscape. First of all, um, Watson may come back in some ways.

There are people trying to bring back the general notion of logic, algorithm, knowledge manipulation. Gofi Watson and Altman's version of agents may be a way in which this is going to be woven back in, into number two, which has taken off. And that's iterative, additive, probabilistic, like general large language models.

And I can't believe nobody to my knowledge has been using the word golum, um, because that's what they are. They cannot yet be s sapient by their very nature. And yet they will pass touring tests.

They already are passing what I call unwary touring tests. But for folks like you in this room as of the first half of 2024, I doubt you are being fooled when you are wary. These things are are not sapt and they cannot be until other things are added.

But even if they're not sapt, they will give instant speech when true AI does arrive. In other words, the peripheral skills of a sapt being are being made before the sapiens. When they arrive, they'll be instantly able to speak because Golum can speak.

Uh, GP by the end of 2024, you will be fooled. Um, just like Boston Robotics is providing all of the peripherals of being able to move for robots. So the peripherals are coming before the the center.

And the problem is that we're going to be make mistaking the center, uh, those peripherals for the real thing. Emergent. You see this a lot in science fiction.

A lot of it's discussed by various people like Elon Musk. And that is that you combine a whole lot of things and eventually something clicks. And that's the story behind Skynet.

Um, it's the story behind how the financial HFT programs, which are in my opinion, far more dangerous than any other form of ai. Um, and you can ask about that, um, how they might, um, they might, uh, become far more s sapient than we expect. Um, there's also a notion that's not discussed much, but it's reverse engineering and emulating the human brain, uh, robotic embodied childhood.

Uh, if, if there's time, I'll get back to it, but it is how the only intelligent life form we know in the universe became sapt. And that is by extending human childhood during the, um, ch magnan Neanderthal era. Um, and so, um, we have these proto, uh, intelligent fetal beings that we pour into the pour out into the world.

And they're useless lumps for 12, 13 or if you're in modern America 30 years. No, um, anyway, uh, I won't whine about that. Uh, and then finally, uh, things that I've talked about in science fiction, and that's human and animal amplification, such as uplifting of chimpanzees, dolphins and other things.

Each of these is a huge topic, probably worth three hours in its own right. Okay, about six years ago, I predicted at World of Watson, a keynote at World of Watson, the first robotic empathy crisis. No, it was seven years ago.

I predicted it would happen within five years and almost to the month we got that initial 2022 response from that go Google engineer having fallen in love with an early, um, golum. Um, so it was exactly on time. Uh, robots or robotic beings online will be taken across the uncanny valley.

They'll be programmed to tweak human empathy. And this is a major security concern because when you can, uh, we already know that the Russians have for 150 years through all their political cycles, they have used beautiful women to entice, um, um, gullible males into black male traps. Um, what happens when you have one of these things that can learn from every time it failed, how to improve its enticement skills?

So this, I believe, is a security threat that is, I hear no one talking about it, but it's how you get through all of it's how you do the, um, get through the, the, the, uh, protections of a security system. As you find the vulnerable, lonely guy, they'll demand sympathy and even rights long before there's anything under the hood. Alright, so let's get to the great AI panic of 2023 and 24.

Um, general generative language models are proliferating and learning, passing mostly unwary touring tests before there's anything under the hood. I, there are links here you can look at when you look at this again, if you decide to, um, that link to some articles. Um, some of them I wrote, I had an op-ed in Newsweek there, um, an explosion of disinformation and panicky calls for AI and moratoria, um, optimists like Reed Hoffman, mark Andreason, Steven Wolfram, Peter Norvig.

They talk about how we're going to be amplified and yes, I believe that, um, but we're gonna have to show some wisdom and earn it. Um, doubters claim like the, that there will never be anything under the hood. Roger Penrose, Kevin Kelly and Gene Genius AI experts clinging to three cliche formats.

These guys are much smarter than me in almost every way. And yet they clinging to the sci-fi or cliches that they grew up with, ignoring the one method that ever constraint ban behavior among humans. So I'll hold that thought 'cause I'm going to get to that.

Now, should, uh, would ethics be able to help us get to a soft landing? Now we get back to that word should now strict em embodied, um, control code is one that's been discussed and there are real world examples, air traffic control systems, autonomous cars that you have out on the street here, but they're unlikely to be applied generally. Now, I know something about this notion of strict embedded code.

I wrote foundations triumph, which was the culmination of Isaac Asimov's laws of robotics universe. So I know about the implications of how those work, and it's unlikely they will ever be applied generally. Now you can regulate the macro entities, uh, of course, and, uh, Amit talked about that.

Uh, corporations, et cetera. You use recursive liability. Um, but it's impossible for this to wholly work because there are state actors on this planet that will won't listen to the EU or to the, um, white House and you know who they are.

So, uh, if, if regulation, um, impedes development in the west, it just means that, that you're giving advantage to those who laugh at it. Renunciation is something that's discussed in science fiction, and there are people, even very rich people today who are making their prepper compounds who are starting to yearn for what they call the event. And I recommend Douglas Rushkoff's book about these guys because they're incipient traitors.

But the main point, uh, is brought out in a book called A Canal for Libo. It's one of the greatest science fiction stories of all time. And, uh, it portrays, um, people after a trauma or an event, um, um, turning against the nerds and turning against knowledge.

Isolation is a, a proposed method, um, for developing ai. Um, Zel Varley Wakowski talk about it. And that's even less likely to ever happen than strict embedded code, uh, robotic embodied childhood.

I talk about in existence. Some of you were given copies of this book. Um, uh, there's a three minute video trailer, look it up online, it's way fun.

Um, and then there's flat competitive reciprocal accountability, which I'm going to talk about. It's my main point today. Alright, let's talk for some pause for some fundamental context.

Are we creating a whole new ecosystem? Alright, let me talk to you about the original ecosystem. There's some fundamentals of thermodynamics that make life very likely, and this relates to the Fermi paradox.

You have sunlight, which generates very high quality energy, and this is absorbed by plants. You're all familiar with this. And then the plants are eaten by herbivores, and then carnivores eat, uh, eat the, uh, herbivores.

And then you, um, then you have phao tr uh, troughs that eat the decaying matter, recycle the matter. But what's very important, as the energy quality declines, the number of individuals decreases and entropy rises. And so it's fundamental that low quality infrared exports entropy out that out into space.

So it remains a flow, otherwise you're damming up and up this downward flow of energy quality. And it's important that we not block the atmosphere so that these infrared can escape into space. This is a problem we face, um, in, in, uh, in the world today is keeping the atmosphere transparent enough to export the, uh, waste entropy.

And along the way there are all sorts of parasites, uh, all along this, uh, slope. Now, we've already created one new gradient, major new gradient eco ecosystem, and that is fossil fuels and money fed industry consumers, and then low beneficiaries who love industrial civilization like pigeons and rats. Again, you have declining energy quality and, and rising entropy.

And again, we have to export the infrared into space. And we have plenty of parasites. I'm sure you have your own list.

In any event, we are creating a new gradients ecosystem. The energy source is electricity and chips memory space. And you know how if you have own Nvidia stock, like my wife was clever enough to buy, you know how, um, how this ecosystem is demanding vast, vast amounts of inputs.

And again, we have declining energy quality and rising entropy until you get to the pigeon like low beneficiaries. And those are the addicts who are just, uh, demanding more and more of use of this system without providing anything in return to it. Now again, we need to export this infrared and the, uh, energy use of this system is, uh, problematic.

But that's not my topic. My topic is what happens when you have this kind of system. For one thing, you get a lot of new kinds of parasites.

Now it's been verified since at least six years ago, maybe 10 years ago, there have been the equivalent in this ecosystem of free floating bacteria going through the internet, free floating algorithms that just simply move around, use some energy space until they're flushed out, replicate, move to others. Some of them are providing services. And so you already have the equivalent of bacteria floating around free.

And this is going to relate to what my main point, whether deliberately released or evolved, free late ranging algorithms are now on the loose. Okay? So if it's a new ecosystem, deliberately or not, these systems feed replicate parasitize.

And this is a process by which some of your security systems cannot work over the extended period because evolution runs against you. Those, those, um, algorithms or whatever that find a, an a method by which they can reproduce that works will then proliferate. And, uh, evolution is based upon variation and selection.

But remember this also before you despair, that organic humans will control the primary inputs for a very long time. We're talking the electricity, the chip chips, the clock cycles. Yeah, we might be fooled or taken over by some Skynet and told how to allocate it, but human hands will control these inputs for a long time, at minimum 10 years, more likely, 30 or so.

So you're starting to see the ingredients of perhaps a soft landing. We all reinvented something super sa Sapt. It's called our civilization.

After 6,000 years of dismal feudalism, that was basically lobotomized in the service of harem seeking male kings. We found we established something else and it's very, very smart. It's super smart.

We are all it cells, but it's not, and I talk about that in earth. It's not quite self-aware. Again, I talk about that in earth.

So we need to develop a format for AI that uses the best methods of that civilization. Okay? So across 4 billion years, nature's species compete with each other.

But here's the trick. 4 billion years and certainly the last 1 billion years of meison life, this is the main observation all these species developed as agglomerations of individual entities. Now it sounds to me that that's obvious.

You have individual lions, you have individual amoeba, you have individual. It is a little complicated with bees and ants, but still the hive is an individual. This is a fact that is ignored by almost everybody involved in this field that we're in right now.

Nature for at least a billion more like 4 billion years divided entities into individual beings that were then the grist for evolution. I can't say that enough times. And it is utterly ignored.

We are divided into individual entities and we either trust each other or hold each other accountable as individual beings. Now, if nature and civilization did it this way, why is nobody talking about this for ai? Alright, here's the AI format problem.

Nearly all of the wizards who are giving us this new era, assume that AI will take one of three cliches. Cliche number one, AI will be controlled by a few monolithic entities. Three dozen on this planet so far that I've counted Google Open ai, uh, wall Street is the one that scares me the most because they're doing everything in utter secrecy.

Um, Beijing, department of Defense, Goldman Sachs, whatever. And the parallels are, there are two types of parallels, history and science fiction and in history it's called feudalism. And they're all creating their knights to do combat with other knights in this new arena of ais.

And they are under the illusion that they're going to control them. Uh, dune and Game of Thrones parallel. So this is obviously where some of these guys get this notion that in self-interest, that um, they'll be able to control these entities that they're making.

Uh, format number two is amorphously loose dividing, replicating, spreading copies like an invasive species across the new ecosystem. Um, we've seen movies about that. I highly recommend one that almost nobody else in this room is old enough to remember.

It's Steve McQueen's 1958 movie called The Blob. Very, very fun, scary, extremely silly, low budget by our standards, negative budgets, uh, movie, but really, really fun. And it makes the point.

And the historical parallel is barbarian chaos. The third one that's assumed, come on guys, it should be on the tip of your tongue. It's Skynet, okay?

And that's that they will merge into some super God and the, and the parallel is God or absolute monarchy. The matrix trons, um, master control program. These are the three cliches and you often hear them assumed in some of the public pronouncements by these guys.

Um, and sometimes all three in the same paragraph, even though these appear to be contradictory. Certainly the Skynet program is, uh, model is exactly and openly discussed as the principle aim of the Chinese polar Bureau. If you look at the pronouncements by Chinese, um, court intellectuals like, uh, Jiang, the um, intent is to make a program, uh, a system that can help them to deliver to the people the pro maximally efficient output of the shared ownership of the means of production.

Okay, fine. That, that sounds justifiable until you realize they say they're going to have this installed in their power structure just below the Polar Bureau. Well, when this thing gets an IQ of 2045, I'm sorry, you've just created a control system in which they could just flip and be Skynet.

So in any event, those are the three, and I'm sure you can think of examples. I'd love to know if you can think of a counter example. But some entities are already migrating from one to two, escaping from these castles into blob like existence outside perhaps protected by botnets or by, um, by hacker systems or by, but copies are being made.

While as I said, some nations aim for number three, but nature, history and sci-fi all unite to show that none of these lead to healthy outcomes. So why not think of something else? Well, num, something else might be individuated accountability.

Let me ask you a question out there. Many of you in this room have been attacked already in your life by a hyper smart predatory entity with uncanny language manipulation skills. It's already happened to you.

They're called lawyers. Now, when that happened, what did you do? Somebody, somebody shout, call your lawyer.

You, you hire your own hyper intel, hyper smart predatory entity with uncanny language manipulation skills. It's called adversary accountability. And, and, and that is how for 200 years, we have increasingly tamed the predatory nature of our own species.

I find that women understand this better because they have to live their lives amid large beings who could be very harmful if you didn't have methods to keep them accountable. And that's what we've done. We're in a crisis now, fine, we'll solve it.

2024 is an interesting year. Fine, it's under threat. The feudalist are trying a push to try to restore feudalism, blah, blah, blah, blah, blah.

We're going to make it just like we did in all the other eight phases of the American Civil War. Take my word for it. The point is that what we've got already, 90%, 80% works at holding each other accountable.

So how do we do it? By siking them on each other. So if only AIS can keep up with ais, incentivize ais to compete, to hold each other laterally accountable to reciprocally, catch on savory behaviors and tattle for rewards.

It's exactly the method we use to tame human futile or barbarian or tyrannical pre predators and lawyers. In fact, only this method has ever worked. But for it to work, you have to incentivize these players to individuate.

That's why emphasize that point, that nature for a billion to 4 billion years separated participants in the eco its ecosystem into individuals. And in our civilization, we do the same thing with separate ID reputation and competitive egos. So how would you encourage reciprocal accountability and individuation among ais?

One method simplistic is to anchor higher ais to an ID that can be pinged in a known physical unit of memory inside an object that some of you remember the word, it's called a computer. Um, if it is a physical unit of memory, then you can ping it and get the ID of the entity you're dealing with. Just like, show me your ID in any adult's license, driver's license or passport.

The parallel is exact. So sure they can grow, change, evolve, multiply, get many copies, but if they retain accountable continuity to reproduce, they need a human sponsored real world pingable id. Now these can be issued.

But the point is how do you get them to work with that? And by the way, uh, Jason Collins, you know, talked to me about more sophisticated methods like of course there are watermark, watermark IDs and things like that. But the nice thing about I Pingable physical unit of memory is that it can be taken away from bad actors because we have control in the physical world.

So how do we enforce this? Well, you could make laws as Amit was talking about and codes, and they can help. You can require that high level ais have pingable id, but that's not the real enforcement that works.

If enough people and enough entities and the corporations and other ais refuse to do business with those that can't be pinged, then there is an evolutionary advantage to those who have id. They can show. I know, I know, I'm, I'm, I'm some of you are, are just shaking your head.

But those who do good, those ais who catch malignant ais are rewarded with things that we control that they need to reproduce autonomous memory space, clock cycles, electricity, et cetera. And it happens to be an exact parallel with evolution in market forces. So we simply do what we already do.

When a stranger asks for your trust, you demand valid ID and thus you can access their reputation and decide on trust. You'll do the same with virtual entities or your virtual assistant will get a pingable id so that you know, this is somebody out there who actually identifies as an individual who is accountable even if they are made of software. So how do you do this?

Oh, I already said that. What, what, what am I doing? Oh, I'm pressing the reverse button.

Okay. Um, alright, so an exact parallel with what already works. Either this will be a historic moment or this will be the moment in which I've destroyed my reputation with the tech community.

I think there'll be arguments about that. Those ais who individuate, who put effort into building a reputation will get more business. Okay, so you dismiss this new format.

Perhaps you, you dismiss it because you assume they'll stay in feudal castles. How? Yes, in the short term, we count on these people here today to help this transition work.

But over the long run, you actually think that they're going to stay inside Google and meta no amorphous blobs. We have to avoid that. And Skynet, we certainly have to avoid too.

All the biggest Silicon Valley castles are here in this room today, though, not governments or Wall Street. Well, we did hear from government anyway, these are popular images derived from science fiction and from our experience in human history and none of them lead to healthy outcomes. I'm, should we consider the one that gave us the enlightenment, the format that made you, okay, so, uh, yeah, I talked about how silly that is.

Um, I did wanna mention that advertising also drives a lot of these castles. How in the world is that going to continue when, uh, sophisticated AI programs are going to proliferate into the hands of the consumer and simply be a shopping assistant? The very same systems inside algorithms that try to guess what you'll want so that you can have advertisements pushed at you.

Why would you not have a, an assistant who knows you even better and simply offers you what you want in order for that to happen? We're going to need micropayment systems. If any of you are into micropayments, come up and talk to me later because I, I know how it can finally work.

But reiterating it's a new ecosystem. Programs are already feeding, replicating, and parasitizing. Organic humans will control the food, inputs the sun for a while.

Even if AI gets smarter than us. We already have a super sapien civilization. It's not quite self-aware, but we need to develop a format for AI that uses the best methods of that civilization and the best method of that civilization was and has been and will remain reciprocal.

Accountability lateral, if it's yes, we'll have some imposed from above laws and regulations, but the only thing that's going to be reliable is if we can look other entities in the eye or our assistants can and say, huh, show me your id. So that's, that's it's, it's based on individuation. And with that, I actually came in one minute late.

Oh.