A must read. Xzistor LAB brain hacker and roboticist Anomalous shares his view on the future of AI.
Anomalous, “AI for Peace – A Hacker’s Dream”
When I recently read in the media of the establishment of worldwide intervention and advocacy groups OpenAI and the Foundation for Responsible Robotics, calling for the control of the development of ‘artificial intelligence’ (or AI in short) to ensure a good outcome for humans, I thought this was a good thing. Impressed by the insight and foresight of these individuals, I have decided to add my pennyworths by contextualising my own research and showing why it is of particular relevance – specifically in terms of this new threat to mankind – and to share my dream for the future.
Yes, we as the AI fraternity perhaps need to admit that we have not made as much progress as we could have made over the past few decades – this in spite of staggering amounts of money having been injected into AI and many groups around the world working on the problem fulltime. What was supposed to be the Robot Age is now often referred to as the ‘AI winter’…
Why did we not achieve the original goal of creating ‘artificial intelligence’ – of understanding and mimicking the human brain? Where are the robots that were supposed to be helping us in every home? Why do we not see machines with human-like intelligence and emotions around us on the streets?
Where are the worlds of Arthur C. Clarke and Asimov?
Luckily my research was spared the chill of the AI winter as I clocked some late nights up in my cosy ‘hacker attic’ also known as…the Sage’s Cave. The discoveries I had made here have been every bit as exciting as the classic old science fiction writers had predicted – and better. My research had revealed to me the deepest secrets of the mind – allowing me to understand how the brain really works and how it can be modelled in a simple way. It also provided a basic framework for building simulations and machines with human-like emotions and intelligence, which proved it was possible to build artificial agents that really do think and feel like humans. When I looked at ‘Simmy’, one of the simulated agents controlled by my brain model (see www.xzistor.com), as it navigated within its learning confine, I did not see just another robot dog or a fluffy pup cycling through its ‘look-alive’ code. This was an ‘aware’ virtual creature desperate to make sense of its learning confine and feeling fear, pain and emotion all the time – even crying for help as it desperately searched for food. The manner in which this model opened up the different aspects of the human an animal minds to me was surprising, and I realised its power as it started to confirm to me the mechanics underpinning numerous unexplained brain states – creativity, thinking, feeling good, feeling bad, depression, euphoria, addiction, intuition, phobia, speech, recognition, dreaming, forgetting… the list goes on. But it offered more than that. It also showed me how society’s current basis for social norms and ethics can be put on a better scientific footing, and how our rational world view can be altered – leading to a less doctrinaire and wizened society. It also made me realise where the world could be headed with AI – one of two roads with two distinctly different outcomes. This is what I want to discuss next.
The aim is not to indulge here in fearmongering or elaborate technical discussions about the brain, but rather to provide my view on the very important notion that a threshold exists, a tipping point which some call the Singularity, when mankind will suddenly be at the mercy of a superior form of intelligence it had created.
I want to take you on a journey with me into the future, for which it will be best to start at the beginning.
The brain has always been one of the greatest mysteries – bedazzling us for ages with its enigmatic powers and seemingly unfathomable complexity. The whole quest of AI was to understand the brain and replicate it in one way or another. This was the original goal of AI.
There were many efforts to conquer the secrets of the brain over the years – like I mentioned vast amounts of money had been invested by governments and private institutions to realise this dream, and promises were made to funders and the public of what will be achieved within set timescales, typically 5-year and 10-year windows. These deadlines were never met.
Within the AI fraternity we were starting to realise that we were up against the most complex biological structure known to man and hope started to fade. Imaginative new projects had to be invented to keep funders interested and researchers inspired – there were swarm technology, human-machine interface, bipedal motion, artificial neural networks (ANNs), rat brains, cat brains, flexible LCD screens, driverless cars, facial recognition, drones, and more recently…deep learning.
More and more groups moved away from confronting the original quest of AI head-on. Today we have robots packed in futuristic-looking plastic shells that can run around the room but won’t know if they were standing in the rain. Like a well-known AI researcher recently admitted: ‘We don’t even know how the roundworm’s brain works which only has 302 neurons’.
Many now fear the AI Winter will turn into the AI Ice Age.
But I know the AI winter is over – forever. The evidence is lying on my table. I know that the working of the mammalian brain can be explained and modelled and I have discovered the fundamental building blocks that, when correctly assembled, produce all the thoughts and emotions we find so intriguing. And building a superior brain using this model is easy. The biological brain is limited in that we cannot really get under the hood and tweak it. With artificial brains we can add many enhanced features – add senses (drag and drop using our ‘wizard’ software), add many more reflexes, combine lateral thinking with topic-focused thinking (at 100 million times the speed of the human brain), link up brains to work as one, plug all these brains into the Internet (later – not now), give access to each artificial brain to the learning of all the other learning artificial brains in the world (easy), run them off one ‘brain complex’ on the cloud, synthesise it all using some of the simplest ANNs…let them be born with the instant wisdom of all other artificial brains that had ever existed…
If you are one of those thinking the Singularity is not real – I want you to think again. But I will propose a concept solution here to show you how together we, and every mortal on this planet keen on seeing Mars colonised, can prevent the Singularity. So KEEP CALM – AND READ ON.
The first question you might ask is how close other AI researchers are to building superior intelligence. And why have others, so much better qualified than me, not found this ‘brain model’ solution.
Answering this question turned out to be every bit as interesting as solving the brain itself. For years I keenly followed the work of my fellow AI aficionados, thinking that, on some fine day, they will also experience that ‘Aha!’-moment when all will just fall into place. To my surprise I saw that this was not going to be the case. Those specifically gunning for the biological brain (not talking about the machine learning folks here!) were either searching too low-level, simply just connecting up more artificial neurons without the right architecture, or they were too high-level, confusing brain ‘effects’ with brain ‘functions’. My brain model helped me to understand the researchers themselves – why their brains were so biased toward detail and why they were unable to escape their own subjective realms. I realised that even if you placed the solution right in front of their noses they would probably not recognise it as such – most would just turn away and keep on piling on more neurons, sticking to the set ways of ‘conventional AI’ (no disrespect guys, but really, what do we have to show?).
The brain is actually a very simple machine, constantly running the same simple algorithms from birth to death, creating its beautiful complexity in the same way a complex cave structure is formed by the simple recurring drips of lime-rich water over time.
My brain model is so simple it can be summarised in a simple diagram on an A4-sheet of paper, and yet in principle it provides all the mental states I can possibly think of no matter how complex or subtle – I was unable to improve on it (a fact that both surprised and delighted me) since I patented it in 2000 and now believe there is no exception to its generality. There is one basic solution and this is it – the Bohr-atom of the mind (my and my dog’s view!)
The point I am trying to make is that if someone else finds the solution, it will have to be pretty much the identical model, else I just can’t see it working properly. Whether it is a brain for a well-adapted fully functional roundworm or a cloud-based data-gobbling Deep Learning Uberbot – the basic structure of this cognitive architecture will simply have to be present in its brain DNA.
And this should be helpful. Because it is not an Uberbot’s ability to search through trillions of representations or potential solutions that will annihilate humans, but its motivation circuitry driving its ‘needs’ and later life ‘emergent biases’ – the working of which is very clearly described by this model. This means when we build the robot we can choose certain settings which will make the robot decide who it will like and dislike.
Great! So we can build robots that like humans! Are we sorted against the Singularity then?
Afraid not.
It all depends on who gets there next or in whose hands this technology falls. We already said the solution is simpler than most people think. And we have seen how ill-willed hackers are hell-bent on pursuing self-interest in an openly criminal fashion. They are not stupid. They know full-well that this IP holds the key to what some believe will be the last trillion dollar industry on the planet and of immense strategic value to those who wants to lead the Robot Race. I have reservations about going completely open source with AI and specifically my model in an uncontrolled way (see my proposal in the final summary below). At the Xzistor LAB (see www.xzistor.com) we are hard-coding safeguards and operating limits into the cognitive architectures derived from this model before allowing the public access to build robots, and limiting all applications to serial (digital) instantiations, drastically limiting its association-storing and processing capability. We also hide KIEs (key innovative elements) to protect our IP and website (instead of further patenting).
So how can we control the worldwide development of AI?
Ta-ra! My daytime job to the rescue! I work in the nuclear industry – an industry which is heavily regulated (this is a bit dry but we will get to my dream soon – ok?). Just like some brave individuals in the AI industry are now risking ridicule – Dwight D. Eisenhower rose before the United Nations General Assembly on 8 Dec 1953 and spelled out the dangers of a potential nuclear apocalypse and called for international cooperation and control. The treaties and conventions that evolved, along with the international advisory body – the IAEA (International Atomic Energy Agency), put global nuclear developments under non-proliferation control and international law. The multilateral nuclear control process can almost verbatim be adopted for the AI industry. Oh, come on – you want everybody in the world developing AI to be regulated? Nuclear is far more dangerous!
Eh…nope.
To date we have always been able to identify and address nuclear non-proliferation transgressions, and we have been able to limit or restore all nuclear accidents and incidents.
The Singularity is by definition unrecoverable.
To conclude on what we can learn from the nuclear industry (and others like the bio-chemical, pharmaceutical, etc.) let’s look at regulation. Take the UK nuclear industry for instance where they have a non-prescriptive regulatory regime. This means to get a licence to build any kind of nuclear facility, the responsibility is on you as the owner to prove (in your own way) how it will be adequately safe. Attached to the licence is a set of general conditions where you have to address numerous common-sense safety-related aspects and show that your design will be safe – now and in future. An important part of a licence application is the submission of a ‘safety case’ comprising a documented, logically-structured set of safety claims, arguments and evidence (basically all necessary safety analyses and results) to prove you understand the safety risks associated with the plant and that you can ensure it remains adequately safe at all times.
The principle is very simple. If the owner of the technology cannot convince a regulator (including industry safety experts) it can make an acceptable evidence-based safety case for the safe construction and use of the technology, no licence is issued and it becomes illegal to continue with the development.
The same concept can be applied to the use of AI, specifically where used in LAWS (lethal autonomous weapons systems) where the safety case will have to show adequate safeguards, margins, human overrides and defence-in-depth in terms of ensuring the weapon operates reliably (this is key) within the boundaries of the combat zone.
Phew! Now for my dream.
Ok, this is going to be a bit Martin Luther King-ish, but heck, the guy did stand up and very boldly spell out a clear vision of a future society where a certain sector of society will have a better life. I want to spell out a vison of a future society where all of mankind will have…a ‘fantastic’ life!
I see cities in space, spreading out from their orbits around Earth in huge flukes like the blades of a fan where people will live and work and play one day. There will be no disputes over land, because there will be enough space for everyone. There will be no fighting over energy, since there will be fusion and solar plants to provide all the energy and oxygen that is needed. There will be no fighting over food as this will simply be shipped from Earth or ‘printed’ in food factories (ok – this is pushing it). The components to build the cities will be manufactured in space factories from materials mined on other planets and the moons of other planets (and from harvested asteroids – wow, I am enjoying this!). The same water will be purified and recycled within each city. There will be no hungry or homeless or marginalised people since everyone will have a respectable dwelling and enough food and also be part of a diverse duty team, and the vulnerable will be protected. Waste will mostly be recycled and the rest disposed of by sending it to the Sun. Transport will be lightning fast and free. Imagine, there will not be a single gun in the space cities, because any breach in the containment will be catastrophic (think vacuum). Robots of all forms will be respected for what they are and happily work side-by-side with humans – helping mankind build spacecraft and cities, colonising planets and moons, building space factories and helping man make the most amazing discoveries. Life will be about innovation, creation and fun!
Surrounded by these majestic space cities, the Earth will still be spinning quietly in the night sky, the ‘cradle of mankind’, perfectly restored to its former beauty – now an UNESCO site and the favourite tourist destination in the whole of the solar system. It now serves as a living recreational park which millions of people visit for holidays, sight-seeing trips and to live amongst the ‘culture colonies’ where all the cultures of the world had been rekindled as living displays, for all to learn from and to enjoy.
This could be our future.
Mankind’s evolution on this planet has been a long and torturous journey. Our ancestors had come a long way from axe-wielding apes to space-travellers, and suffered severe hardship and pain along the way. Many paid with their lives. Those who survived fell victim to drawn-out wars and bloody conflicts, poverty and suppression, enslavement and indignation, illness and starvation, brutality and rejection, natural disasters and calamities. Men, women and children through the ages have braved times of unspeakable misery and suffering – to get us where we are today.
We cannot let it stop here.
We owe it to our forebears to develop AI responsibly and pass it on to the next generation. As a species we have survived by using our minds and powerful problem-solving abilities – and the scientific progress we have made over the ages has been nothing short of phenomenal. The next generation will face new challenges, and like us, they will need to take the lessons from the past, and build on that – and use the foundations we pass on to them, to ensure nations can unite and collaborate in humanity’s quest for survival. We need to make sure the incalculably high price we have paid to reach this point in our evolution, was not in vain…
Thank you.
So to summarise please find a proposed action list below:
- Awareness campaigns: Everybody helps to fuel the discussion now!
- OpenAI and Foundation for Responsible Robotics yes, but suggest with a funded Blue Sky Lab to ‘ring-fence’ break-through technologies
- Create a safe haven for development of all the safe and beneficial AI areas
- Stay ahead on the potentially high-risk AI areas – maybe start with LAWS
- Robot rights – if you can model the human brain and use it to build robots, you absolutely need to consider the subjective states of the robot too – pain, fear, stress, etc. Please bring this into our ethics discussions (we don’t want one day to be accused of cyber slavery!). This means Asimov’s Laws need to be changed as we have started to do here:
- International advisory body – International Artificial Intelligence Agency (IAIA) created with a clear charter grown from the Blue Sky Lab’s Ethical Design and Development Code (EDDC). Offering guidelines, advice, safety standards – United Nations (UN) linked.
- Start looking at Nuclear Non-proliferation Control (and others) as exemplars for AI international legal framework – UN linked.
- Start looking at Nuclear Industry (and others) as exemplars for member state AI national regulation – UN linked.
- International AI symposia and awareness campaigns. Voluntary registration of international ANNs-based AI development projects in UN member states with IAIA.
- Treaties, conventions, etc. under international law – with enforcement arrangements signed off – UN linked.
- Drive to formalise member state national regulation of ANNs-based AI development projects. Probably need a national AI Act calling for an AI regulator.
- International expert intervention teams to identify and counter unsafe and unethical AI practices.
- As AI risk increases – legal and enforcement structures can be expanded.
(Well done to all who are standing up for Ethical Robotics – I salute you! If you have questions about my model, check back from time to time to the Xzistor LAB at www.xzistor.com or send an email to xzistor@live.com and look for posts about the X-zistor Concept brain model in our free public area. Here is a short intro video about the lab: Intro Video. I hope you will find it interesting!)
Kind regards,
Anomalous
