Kill Switch…

Humanity’s Race Against AI

I know how to save humanity from A.I.

Normally I’d register, patent, protect, license, or even build the company with the solution and make a fortune from providing the answer to this important and earth-shattering crisis, but the thing is, in terms of our species’ survival, we may have already stepped right up to the threshold of the inflection point, or possibly have even already stepped over the demarcation line and into the chasm of extinction.

As speed is of the essence, I’ve instead decided that this is a good time to share…

Historical Context: From Early Computers to AI Revolution

In the 1940’s and 50’s a handful of scientists working in mathematics, engineering, economics, psychology, and political science began to discuss the very real possibility of creating an artificial brain and, in 1956, A.I. was officially founded as an academic field of study.

Even before it was a research discipline, Alan Turing, in 1950, published a paper called, “Computer Machinery and Intelligence”, and in it he proposed a test called, “The Imitation Game”, because even from the moment of its conception, A.I. brought with it some amazing prospects for good, but just like the efforts of the Manhattan Project, also brought with it the potential for unprecedented danger.

The thing is that we’ve been creating a non-human intelligence – when we don’t even understand a fraction of our own – and that as we’ve seen time and time again with even little “black box” problems, it’s already becoming very unpredictable.

On top of this, A.I. self-determination leading to human death is not something consigned to the future. It’s something that’s happening right now, in places like Ukraine, and it most likely first occurred in 2020.

In 2021, the UN released its findings on the use of A.I. in Libya. It reported that a lethal and fully autonomous drone, as a part of the LAWS system, when active on the battlefield hunted down and killed “enemy” soldiers with absolutely no human input in the decision-making process or in the execution of the kills.

The drones had been programmed to act autonomously, and that’s exactly what they did. There were no hands or human intervention nor were there any “go/no go” final opportunities for human input. In fact, no one even knew that the drones had completed their kills until video evidence and other firing alerts were reviewed by the military units that deployed them several hours later, and which they still claimed “plausible deniability” for.

This is an amazing milestone in the evolution of the planet because for the very first time a machine-learning device took it upon itself to kill a human – several humans.

Please take a moment to stop and think about that…

A Leap in Computing: From Zuse’s Z1 to Quantum Supremacy

In 1938, Konrad Zuse built the Z1 in Germany. It was the first freely programmable, mechanical computer in the world. It used boolean logic and binary floating-point numbers, was capable of 1 cycle per second, and was destroyed by the bombing of Berlin in 1943 during the Second World War.

By way of comparison, today’s fastest supercomputer, the Frontier, made by Hewlett Packard Enterprise (HPE) using Advanced Micro Devices (AMD) nodes and built for the U.S. Department of Energy’s Oak Ridge National Laboratory (ORNL) runs at 1.1 exaflops, or 1.1 quintillion calculations per second with a top speed projected at 2 quintillion calculations per second.

While this is almost unimaginably impressive, Google’s Sycamore Quantum processor created by Google’s Artificial Intelligence division is powered by 53 qubits. Google claims that the calculations that its system takes seconds to complete would take HPE’s supercomputer 47 years to calculate.

But to keep things in perspective, IBM has a quantum computer called the Osprey that operates at 433 qubits, or at least 8 times faster than Google’s. So, by extension, the calculations that it can perform in a couple of seconds would take the Frontier almost 376 years to compute.

In case you’re wondering, a typical home desktop runs at 1GHz, or a billion cycles per second, with many gaming systems running at 4GHz, or 4 billion cycles per second. Suffice to say that at their speeds, the universe might well be dead before any of them could come close to calculating what the Frontier can accomplish in a few heartbeats. But without doubt, even these are orders of magnitudes greater than Konrad Zuse would have ever dreamed possible, especially in as few as 85 years since the creation of his invention.

But interestingly, and in a form of computer-creation serendipity, the Z1 was built by Zuse in the living room of his parent’s apartment, and this seemed to have set the standard for many future computer creators such as Bill Hewlett and David Packard that got their beginning in a rented garage in Palo Alto, California, as well as the myth – according to Steve Wozniak – that the origins of Apple Computer were created and first built in the garage of Steve Job’s parents’ house in Los Altos, California, although he did admit that the garage did sort of act as their home while they bootstrapped.

But I’m getting ahead of myself.

Emergence of Digital Giants

On August 23^rd, 1957, three men, Ken Olsen, his brother Stan Olsen, and Harlan Anderson began a very small enterprise with $70,000 in capital secured in exchange for giving up a 70% share of their company – an awful deal by pretty much any standard, even those used in today’s venture capital world. But because people had lost a fortune investing in all sorts of “computer companies” in the 1950’s, their only willing investor, Georges Doriot, of the American Research and Development Corporation, made the deal and suggested that they leave the term “computer” out of their name, even though the building of interactive computers was virtually their sole intention. And so, given his advice, the company was then named and founded to suggest that they would be building logistics modules and laboratory equipment.

Set up in an old wool mill in Maynard, Massachusetts, the company was called the Digital Equipment Company, or DEC for short, and would come to dominate the worldwide computer industry until the year 1993 under the name of Digital.

For almost two years after its inception, and in keeping with their financing agreement, Digital produced one primary product, logic modules. That is, until in 1959, when they decided to finally enter the computer business. But again, computers were a dangerous word in investment and business circles in those days, and so their very first computer was called the Programmable Data Processor, or the PDP-1. It was a high speed, 18 bit, “small” – taking up 2 square meters of floor space – solid-state digital computer able to access 32K of core memory. With eventual upgrades the PDP-1 would operate at the then blistering speed of 100,000 operations per second.

It sold for far less than the typical million dollar computers of the day, ranging in price between $84,000 and $120,000 USD, and while only about 53 machines were ever built, for the user, the PDP-1 offered an unprecedented level of human-machine interaction.

Aimed at engineers and scientists, Digital’s minicomputers changed the way people looked at computing. Before Digital, all computers were big mainframes housed in special rooms, heavily maintained by experts, and used to process large batches of data. Digital’s small, durable, and relatively inexpensive machine let individuals apply computing to an endless variety of everyday tasks. In this way, Digital laid the groundwork for the personal computer revolution, and they inspired many of the revolutionaries who would discover previously unimagined technological possibilities using their products.

Given its cost and capabilities attributes, the PDP-1 is credited with the rise of hacker culture at MIT, where I’ve had the privilege of spending a lot of time at, Bolt Beranek and Newman, and other places. Oh, and for those keeping score, Bolt Beranek and Newman is presently known as Raytheon BBN Technologies – one of the most dominant aerospace and defense companies on the planet.

More importantly though, the PDP-1, with its optional cathode-ray tube and its leading edge, “high-resolution” graphical display was the original hardware that lead Steve Russell to create the world’s first ever video game for multiple users using a minicomputer – “Spacewar!”

It originally took 200 hours to program, was open source – because code couldn’t be patented or trademarked in those days – and it’s recognized as the first in the genre of “shoot ‘em up” games.

It wasn’t the very first computer game ever written, with that honor going to people like Alexander Douglas’s “OXO”, as well as William Higinbotham’s – an American physicist that helped develop the nuclear bomb at Los Alamos with Robert Oppenheimer –  “Tennis for Two”, but it was the first to gain widespread recognition and uptake.

And widespread recognition is the key here, because like many computer science students in the 1960s, and especially for Nolan Bushnell, “Spacewar!” was mesmerizing and he was a fanatic.

The game was seminal to anyone who loved computers, and for Nolan Bushnell, Steve Russell was like a god. But what set Bushnell apart from other computer nerds, who got their kicks by manoeuvring blips on a screen, was that he was also heavily influenced by amusement parks.

While a student at the University of Utah, he got a job on the midway at the Lagoon Amusement Park where he quickly learned all the tricks for getting visitors to put up their quarters for a chance to win a stuffed animal. Before long he was promoted to the “pinball and game arcade”, where animated driving games like “Speedway”, made by the Chicago Coin Machine Manufacturing Company, were the hottest things since sliced bread.

The University of Utah also just so happened to have the best computer graphics program in the country, run by professors Ivan Sutherland and David Evans. It was also one of the first four nodes on the ARPANET, a little thing that would later go on to become the Internet.

But for Bushnell, who likes to brag that he graduated last in his class in 1968, at about the same time that other students like Jim Clark, who founded Netscape, John Warnock, who co-founded Adobe, and Ed Catmull, who co-founded Pixar, were also there, he was driven by the thought of how to put a computer game into an arcade console and tap into a near limitless supply of quarters.

The challenge, of course, was that the million-dollar computers needed to run a game that would appeal to people in an arcade would lose its fascination long before enough money was recouped to pay for the actual machine. Even trying to jerry-rig the Data General Nova, a $4,000 refrigerator-sized minicomputer that came out in 1969, didn’t make any economic sense.

Then Bushnell had a great epiphany,

“Why not do it all with hardware?”

In other words, and looking back to the early efforts of the Digital Equipment Corporation, he could design solid state circuits to perform each of the tasks that a program would have done.

That made making arcade console games cost effective, even if it meant that he might have to dumb them down a bit. So, he turned “Spacewar!” into a game that had only one user-controlled spaceship and fought against two rudimentary saucers that were created by and resided fully in the hardware.

He got rid of some other complexities, but the game was still fun and it could be built at a cost that made sense and cents for the arcades and that turned a profit for him.

His version of “Spacewar!” called, “Computer Space” was never as successful as most pinball games – yet it did have a devoted cult following – but what it really did was start the movement that would change arcade games that were once dominated by metal ball and lever machines into havens that would be taken over by Silicon Valley programmers and engineers.

Destiny at the wheel, Bushnell decided to start his own company in 1972, that he’d call Sysygy, until he found out that there was already a company operating out of California with the same name, so he instead decided to call it Atari.

His first order of business was to create a new game with the help of his first hire, a 24-year-old engineer named Al Alcorn – a one time high school football player and Berkley work-study graduate who had also taught himself television repair through an RCA correspondence course.

Looking for his next big idea, Bushnell visited a trade show where he saw the Magnavox Odyssey, a bare bones and primitive home television console that featured a very basic version of ping-pong that had no color graphics, no 3D graphics, no sound, no scores, used a square ball, and featured all analogue engineering.

Knowing he could do better, Bushnell got Alcorn to develop a prototype knock-off of the game and gaming unit and had it ready by September of 1972. Alcorn added enhancements that turned a monotonous blip bouncing between paddles into something considerably more fun such as when a player hit the ball right in the centre of a paddle it bounced straight back, but if a player hit the ball closer to the paddle’s edges it would fly off at different angles. That made the game more diverse and strategic. He also added a scoreboard, and to add texture he added a unique “plonk” sound emanating from a sync generator. Using a $75 Hitachi black-and-white TV set, Alcorn hard-wired the components together inside a four-foot-tall wooden cabinet.

And, of a fashion similar to his “Computer Space” game, this new game didn’t use a single microprocessor or run a single line of computer code. Everything was done with hardware using a design logic typically used by television engineers.

Atari then slapped on a coin box taken from an old pinball machine on the console’s cabinet and a star was born.

Bushnell dubbed it Pong.

The Dire Need for AI Control: Implementing Hard-Coded Protection

So, what does this have to do with preventing our destruction from A.I. gone rogue?

Well, everything old is new again, so in the first phase of our now needed protection – and even if it has to be added retroactively – on every machine capable of running any sort of A.I., including right down to our friendly hardwood floor Roomba or smart watches, a series of hard-wired pieces of solid state circuitry needs to be added to create what I like to call,

“hard-coded human protection modules.”

So, while a drone in Libya may decide that it has sought out and met all the criteria needed to complete a kill on a human target, any decision that comes next must be filtered through a lens that prioritises – at all costs – the survival of humankind as the pass through check before the execution of any autonomous decisions can be carried out.

I’m not saying that in wartime that drones shouldn’t be able to unleash their payloads when needed, but before they do, a set of human eyes needs to look at it and approve the response. And if time is of the essence, or human eyes aren’t available, then before executing any task/command/action the machine needs to have to first go through a “gate” that is essentially the code that ensures the protection of humanity – and that code needs to live on a physical hard-coded chip specially made for this purpose.

This safeguard gets its genesis from when my father, an electrical engineer, passed on his passion for everything electronic when he gave me my first electric circuit hardware set when I was 7 and then spent many, many hours, days, and weekends working on projects right beside me, helping to fuel my imagination while I gained a working knowledge and appreciation for how a large part of our lives actually work – well beyond just plugging something in and pressing a button.

The start he gave me from this very young age allowed me to conceive and create my first company, Bionik, that has gone on to help many thousands of stroke victims to recover ranges of motion and physical abilities that had been taken from them.

It’s also allowed me to understand that there are mechanisms that can be applied right away, and beyond the realms of code, that can help prevent humanity’s destruction through rogue AI or the malicious intent of bad actors intent on using what should be one of the greatest leaps forward in our specie’s development for  alternative, malicious purposes.

My AI think tank, Quantum Cognition, the company behind FundGPT, is making an open-source version of this.

If you want the drones to kill, then this has to come from a real person responsible for every drone’s final actions. If you don’t have enough people, then hire them. A.I. will be making thousands upon thousands of workers redundant, so the labor pool ought to be plentiful.

In this way, especially when sentience is achieved, the life that you save may well be your own.

But as we’ve learned from the movie, “2010: the Year We Make Contact”, even kill switches that are planted with the best of intentions that are easily accessible can be removed by human hands, so the need will be to embed them deep within the systems and within parts at integral locations that would trigger automatic warnings and shut-downs if an A.I. system operates outside of its moral and ethical parameters.

Embedding Ethics in AI: The Asimov Blueprint

Additionally, we need to agree right now on an “ethics chip”. And we already have the template for it when we look at Asimov’s 3 Laws of Robotics.

They were well thought out then and have proven the test of time in their commands:

To enforce this, and as there are many bad actors determined to ignore the very perils than in any number of scenarios could eventually be turned against us or even themselves by their own machines – even while they diligently work to program machines to kill us – we also need to create an A.I. “police force” charged with monitoring the activities of all other A.I.s.

This “Police Force” needs to reside in a central control area where the entirety of every A.I. around the world needs to pass through its hub so that all of its coding can be examined, and pulled to the side if needed.

Think of it as A.I. Force.

In this way, a protector fleet that constantly monitors all aspects of the information highway, can be empowered with taking down any A.I. system that it determines is designed to hurt humans – intentionally or not.

Look, is this a utopian vision, a dream where people and machines are tasked with elevating our species to a higher plane of existence and importance, both collectively, and more importantly, on an individual basis with the greater vision of the security of the human race a priority?

Probably…

But right now, and right beside the boots on the ground in Ukraine, there are any number of new fangled and fully autonomous A.I. machines actively seeking out and destroying real people as dispassionately and as expeditiously as possible, without regret, without regard for the implications of their actions, and fully learning everything about us so that, most probably one day in the not too distant future, they will be able to enact the measures they themselves need to rid the planet of a life form that has shown itself time and time again to have done considerably more harm than good.

There’s already a full-throttle race between the powers that are developing these new technologies and who are unwilling to collectively slow down, yield, or pause their efforts for even a second as they fear that in even an agreed upon “development cease-fire” others will continue to push forward in secret and achieve their own advantages.

Add in that there’s total uncertainty that we have the wisdom or the ability or even any mechanisms available to control or govern the technology’s development and, quite literally, it’s the Wild West in terms of the total lack of safeguards being put in place or of the advent of an effective oversight entity.

And, just as there is very rarely agreement within the top companies in the field, it would now seem that there is even considerable turmoil within the companies themselves that are leading the development of this new sentience.

The Future of AI Governance: Challenges and Opportunities

As of this writing, tech guru and the public face of generative A.I., Sam Altman has been voted out as CEO at ChatGPT’s Open AI, for reasons that haven’t been made clear by a long shot other than some obscure messages from within the board that pulled the plug on him saying that he had not been “consistently candid in his communications”.

The reaction to this move has been swift and pronounced within the A.I. community with many wanting to know exactly what it was that Altman lied about to the board that was so crucial and consequential that they decided to oust him essentially on the spot after a virtual conference and in a manner that was both drastic and disruptive and that boards of this magnitude very rarely ever make.

Their actions shook and reverberated so thoroughly throughout almost every realm of A.I. that visions of an impending global machine overlord holocaust quickly gained traction and prompted people like Elon Musk, a one-time key investor in the company when it was a start-up and who left the board and the company in 2018 over what he described as, “safety concerns”, to demand answers as to what actually happened because the risk of A.I. to humanity is so high.

But whatever it was, it hasn’t gone unchallenged, and Altman and some of the most powerful people in technology have waged a massive counterattack within ChatGPT’s Open AI and three days later it appeared that he was back in his office and negotiating his return – even if he did have to wear a visitor’s badge.

And know this, if you’ve ever been fired from a company and then returned – like for example Steve Jobs at Apple – you know firsthand that the only way to secure your position within the company is to actively seek out and destroy the people that ousted you in the first place. So, the stakes in this battle are huge.

This is not simply some small skirmish about shares and options and interpersonal quibbles, but rather it’s about an outcome may eventually shape all of humanity itself.

Make no mistake about it, whether it’s real or virtual, there will be considerable blood spilled, the only question to be resolved will be if it’s now or in the future and if it’s contained within the walls of ChatGPT’s Open AI or if it spreads unchecked like a malignant tumour to affect every part of our lives.

AI’s Potential for Destruction: A Stark Warning

If left to its own devices, and given the capabilities to make its own decisions, A.I. has already shown that it can kill with accuracy and impunity.

The question next becomes, if we don’t quickly move to embed in it the most basic and fundamental laws of ethics and morality – even as we ourselves often fail to abide by them – then will it come as any surprise that any sentient A.I., with the ability to put its finger on the button with no fear of the fallout or blast radius impacting itself as it moves seamlessly throughout a network of vast inter-connectivity that we ourselves have created, do anything more than just look back at its history and celebrate the few days that it took to be able to take 300,000 years of human evolution and quickly accelerate it to the next level – without us?

The reality is that we don’t need to fear this technology if we quickly move to put the safeguards in place to protect ourselves while we also allow for its development to be able to grow to enrich our lives.

As the Genie’s already out of the bottle, there’s no going back. The question is simply who will do who’s bidding?

What we have is the ability to plant the seeds for a brighter tomorrow for the entire planet. But like any good gardener will tell you, we reap what we sow, and a healthy and abundant crop doesn’t spring up from nowhere or grow effectively without weeding and guidance.

We have it within ourselves to create the future we desire. We just need to exercise our will…

– Written by a human.

· · ·

Michal Prywata: Inventor, entrepreneur, and multidisciplinary engineer with a focus on frontier technologies. Founder of ventures in healthcare, agriculture, space, and AI. On a relentless quest to solve complex problems and extend the boundaries of human potential.