AGI- Artificial General Intelligence - the state of the Art, & will it be the end of mankind?
24 August 2021
“With AI, we are summoning the demon” – Elon Musk, or “Our Final invention” by James Barrat… and Martin Rees, Astronomer Royal, of Cambridge University, UK, and many others, prophesize huge risks or even potential doom for mankind from advances in Artificial Intelligence, (“AI”), or specifically Artificial General Intelligence (“AGI”). Others prophesize huge benefits for mankind, from new medicines, much more leisure time and elimination of arduous or boring jobs and tasks, and the solving of many previously unsolvable problems, that will provide a boon for mankind.
Artificial Intelligence, AI, has been steadily progressing over recent years, (after many decades of “AI winters”) to a point where the launch of GPT-3 in late 2020, has signified a step change towards Artificial General Intelligence, or “AGI”.
This note reviews the state of the art in AGI, various players, and the potential implications, in the following sections:
1) What is Intelligence, and Artificial Intelligence?
2) What is AGI?
3) What can AI actually do at present?
4) Who are the leaders in this technology?
5) Brief history of AI
6) China vs America vs Rest of World
7) What next, and by when?
8) Robotics- The full system- robots with advanced AGI- replacing or augumenting humans?
9) Should we be worried? And AWS, Jobs etc
10) Summary
1) What is Intelligence, and Artificial Intelligence?
General Intelligence could be described as the ability to solve a range of problems, learn and recall, and take effective, human like action in a wide variety of environments. The faster and better one can solve the ever more complex problems, could potentially identify them as more intelligent, assuming they have the right base of knowledge. Knowledge is an intelligence amplifier - so those with the best information / data can also potentially appear to be “more intelligent”. So the arrival of the internet and Google and other search engines would appear to have amplified the collective intelligence on the planet .. and made it much easier for people across the planet to learn things. However there are many definitions of intelligence, and many intelligent experts cannot agree on the definition of intelligence, so this is what I am using here.
Artificial Intelligence, AI, is the use of computers to simulate human Intelligence, and there are also many different definitions of what Artificial Intelligence is. Currently it is mainly “narrow AI” so programs are used to solve a specific “narrow” task, much programmed in the Python programming language. This may include a range of activities currently done today, such as self driving cars, disease mapping, smart assistants, conversational marketing bots, automated financial investing and so on. The next big step is to make such programming, currently based on neural networks, able to effortlessly switch to perform tasks across a wide variety of domains, becoming more “general intelligence”.
On a human level, there are a multitude of intelligence tests, some for specific things eg dementia / Alzheimers, such as MMSE test, and various groups that have clubs you can join if you perform to the required level for entry to the club eg Mensa- IQ > 132 (One in 50); MegaSociety - IQ >176. (One in a million; LAIT entry test); Prometheus Society- IQ > 164; (1 in 30,000); Triple 9 Society – IQ >146 (SB5) & >0.1% of population. And the estimated IQs of some famous people viewed to be very smart:
Einstein: 160
Elon Musk: 155
Bill Gates: 160
Mark Zuckerberg: 152
Steven Hawking: 160
Mozart: 150?
Steven Jobs: 160
But intelligence measurement is not an exact science; and as mentioned there are many different tests, and higher IQ tests are less reliable than those close to the median. And how does one accommodate the situation where one person may be a world leading genius at one thing, but perhaps poor at the others eg Einstein or Hawking at physics, but perhaps less talented at other disciplines? So IQ can be a very rough and imperfect guide.
But if you must have a basic guide, then the Stanford-Binet scale below is often used.
S-B-5- sub tests assess five cognitive factors: Fluid Reasoning; Knowledge; Quantitative; Visual–Spatial; & Working Memory. Other tests include WAIS-III/IV/V, Raven's APM, GMAT, MAT, Wechsler, HAWIE, and many more.
One cannot help but wonder about children that are claimed to read multiple languages before age 5, or graduate college or enter University at age 10 or 11. Or Kim Ung Yong, invited to NASA at age 7… supposedly with an IQ of 210…see 10 Most Genius Child Prodigies in The World - Wonderslist Why would this be? Why is their neural wiring different and can this be emulated with others with different wiring or instruction ore environment? Or Acquired Savant Syndrome- where after a head injury someone without previous aptitude acquires genius level skills in a small area..? Or Elon Musks neuralink company- can we augment a person intelligence with an embedded computer interface… so are there ways to hugely and safely boost the intelligence of humans?
2) What is AGI?
AGI, or Artificial General Intelligence, is where computers can perform a wide range of cognitive tasks that can be done by humans, and learn and adapt to the problem or situations before them and arrive at the correct action(s) or conclusion(s). The programming allows the computer to review situations or data on its own, and come to the correct conclusion. Indeed, the original programmers can have trouble understanding how the computer is arriving at certain decisions, they can become “black boxes”..
“The Singularity” is a term often given to the point where computers become as smart as the worlds smartest person. Given that “smartness” or intelligence can be defined in many different ways and can vary significantly by different fields eg physics, mathematics, chemistry, biology, versus politics, philosophy etc, then the Singularity will be very wide ranging, predicted a number of years ago by the futurist Ray Kurzweil to occur in 2045. Others are predicting an even sooner arrival (and some say it will never happen..) Therefore the Singularity may first occur in a specific field, and then move to encompass many fields, and eventually all fields.
Some worry that whoever achieves this task first could then dominate the planet because the smartest machine could then propagate itself to thousands of other machines and exponentially increase learning, and then indeed plot to prevent anyone catching up, and indeed, if combined with advanced robotics then might decide that humans are not useful or indeed dangerous and only to be kept as tightly controlled pets.. as they will have a demand for resources that could threaten the overlord robots. This is the view of some of the dystopian researchers on AI.
Others believe, probably correctly, that AGI will eventually make large portions of the workforce obsolete. Where previously different and new jobs emerged eg web designer, as new technologies were developed … we are now approaching an inflexion point where if a computer or humanoid can do everything a human can do, where is the need for a less efficient or less capable biological organism.
3) What can AI actually do at present?
As mentioned above, AI today performs a range of activities, such as self driving cars, disease mapping, smart assistants, conversational marketing bot, automated financial investing and so on.
And Open AI, OpenAI , of San Francscio (co-founded by Elon Musk and Sam Altman who pledged $1bn) recently – in 2020- received a $1bn cheque from Microsoft, for further research, and Microsoft getting exclusive licence to the Source Code for GPT-3, (Generative Pre-Training- 3) which they developed, is said to be among one of the most advanced AI systems to date.
Open AI governs the use of GPT-3 to prevent the actions of many potential bad actors, such as using AGI to pretend to be people and have conversations on various chatrooms and therefore sway opinions on a range of subjects, from political views to laws. GPT-3 is aimed at natural language answering of questions, but it can also translate between languages and coherently generate improvised text. OpenAI stated that full version of GPT-3 contains 175 billion parameters. Open AI codex is most capable in Python, but it is also proficient in over a dozen languages including Javascript, Go, Perl, PHP, Ruby, Swift and Typescript and Shell.
There are a range of examples of what GPT-3 can actually do, Examples - OpenAI API, and pricing for its use, to cover things such as your own chatbot, translating difficult text to an easier to understand text, grammar checking, translating python code to human understandable language or explaining code, chatting as though you were chatting with friends, and so on. If you use the internet today you will already be encountering many systems that use AI, and GPT-3 will be ever widening the range of activities you will encounter. And this is only beginning…
And various AI programs have been defeating humans at games such as Chess and Go – this is discussed further in the next section, showing ways to out-think and outcompete humans even in games and activities requiring creativity.
The AlphaFold program from Google/ Deepmind has predicted the structure of nearly the entire human proteome (the full complement of proteins expressed by an organism) and made it available through a public database. The deep-learning neural network, developed by Google’s artificial-intelligence firm DeepMind, predicts the 3D structures of proteins from their amino-acid sequences. AlphaFold has also predicted almost complete proteomes for other organisms, ranging from mice and maize (corn) to the malaria parasite. The more than 350,000 protein structures vary in their accuracy. But researchers say the resource — which is set to grow to 130 million structures by the end of the 2021 — has the potential to revolutionize the life sciences. So this is an example of an AI powered program that has cracked a problem that has previously defeated human minds – stated as “Deepminds protein-folding AI has solved a 50-year-old grand challenge of biology. AlphaFold can predict the shape of proteins to within the width of an atom.” The breakthrough will help scientists design many new drugs and understand disease”, so offering huge potential benefits for mankind.
Indeed Deepmind is also said to saved Google hundreds of millions of dollars per year in electricity cost in their datacenters by calculating the optimal running of their servers, a very real and very profitable application of AI.
4) Who are the leaders in this technology?
America and China are broadly the leaders in developing AGI. That said, Google bought Deepmind, (for some $600m in 2014) which was London, UK based, headed by Demis Hassabis. Demis mission is to “solve intelligence”, and Deepmind is arguably the world leader in developing AGI. He was originally backed by a Silicon Valley luminary Peter Thiel.
When Deepminds algorithms beat the Chinese Go Master Ke Jie- the number One player in the world in Go (after earlier beating the South Korean champion Lee Sedol 4 to 1), there were 280million TV viewers. Seeing this is said to have caused a “Sputnik moment” for the Chinese.
The big companies and leading groups are Google (with Deepmind, and Google Brain); Facebook; Apple; Microsoft; Amazon, Baidu, Huawei, and Open AI, and maybe iflytek, listed on the Hong Kong stock exchange, with a value of some US$15bn. Much revolves around the people and groups, Google has or had Geoff Hinton, Sutskever (moved to Open AI) & Krizhevsky and Hassibis, Legg and Silver and Jeff Dean at Google Brain. Facebook has Yann LeCun. Baidu has Andrew Ng. Open AI has Ian Goodfellow, formerly of Google Brain. Others work in different guises, such as Qi Lu and Yoshua Benigo. Subsections include GANs- Generative Adversarial Networks, where Goodfellow has a strong expertise. Google also developed Tensorflow, a key component of many AI systems.
Some view the leading people in AI as the deep Learning pioneers such as Geoffrey Hinton, Yann Le Cun, and Yoshua Benigo , where all three shared the Turing Award, the most prestigious AI awards, equivalent to a Nobel prize, with these three being also known as the “godfathers of deep Learning.
If you would like to see a long list of leading AI people at various companies and Institutions, see page 321 of the 2021 book Genius Makers.
An AI group, the Israeli AI21 Labs.. in August 2021, showed off a system bigger than GPT-3..
And we cannot know which Labs within companies are developing the next big advances, and we may not know until it is announced..
So who will win the race for the singularity, will it be those with elite researchers leading the field, or those with armies of software coders..? Only time will tell.
5) A [very] Brief history of AI
In the 1950s, a number of engineers and scientist started working on AI, such as Frank Rosenblatt, who developed the Perceptron. An arch rival, Marvin Minsky, was through his critiques and actions able to prevent Rosenblatts work progressing, and AI essentially made little progress until the 80s, despite many high hopes. There were many “Tribes” who took different approaches and looked down on the approaches and philosophies of others. Geoff Hinton (and David Rumlhart) managed to spark somewhat of a revival in the neural network approach pioneered by Rosenblatt, but it was not until 2015 that things began to take off, with the neural network approach taking the lead, and large companies fighting to recruit the best or most experienced brains.
6) China vs America vs Rest Of World, and the first mover advantage
Kai-Fu Lee, founder and former Head of Google China, stated that in 2017 China “had its Sputnik” moment ie when Russia showed it was ahead of the USA in the space race when it was the first to put a person – Yuri Gagarin- in to space with its Sputnik spacecraft, it shocked America and its people and galvanised and inspired the whole nation to drive to win the race to the moon, culminating in the USA being first to put a man on the moon in 1969. The moment of shock, the Sputnik moment, for China was when in 2017 their best Go player – Ke Jie- the number One ranked Go player in the world- was beaten by Googles Deepmind AGI program, Alphago- watched by hundreds of millions of Chinese. Many huge efforts were then sparked in China for them to seek to win the AGI race.
As stated above, a leading company in the race to true AGI is Deepmind, which Google acquired for some $600m in 2014. The company founder Demis Hassabis is still leading the efforts, based in londons King Cross, where his mission is to “solve intelligence”.(In 2019 they had a loss of $649m, so it is an expensive business.., on sales of >$300m), employing over 1,000 people. But the prize is huge.
In 2020, as mentioned above, they announced the solving of a 50 year “huge biology problem”, predicting protein structure / folding.
There is a clear first mover advantage, as the first advanced AI out of the box that can improve itself is already the winner.
So whilst the USA currently leads the pack in AGI, China could gain rapidly, and they have unveiled a plan to become world leader in AI by 2030. China is treating AI as its own Apollo program. Baidu has its own AI Labs, and own Tensorflow - called PaddlePaddle.
7) What next, and by when?
The next thing on the horizon is GPT-4… or something produced by another AI group, such as Israels AI21 Labs.. in August 2021 showing off a system bigger than GPT-3.. And who knows what might come out of the leading Labs of Google, Facebook etc.
The futurist Ray Kurzweil forecast the Singularity would occur in 2045, when computers will be as smart as the smartest human. It can then work on making itself even smarter, and whilst Einstein could only impart his knowledge to a small group of people at a time -those at a level that could understand his claims and discoveries, the program for the “Singularity computer” can be propagated to millions of other computers almost instantaneously. This could lead to an exponential growth in AGI outpacing human capability, and indeed outpacing the nearest competitor, so winner takes all. However some of these capabilities will be well hidden by Government Institutions only to be revealed when the stakes are high - economic or military.
The next war will likely be a different war- possibly largely cyber and AGI driven, possibly biologically driven, and unlike prior wars. Imagine what will happen if all electricity and communications and water and food supply is shut down in a country… rapidly there will be no petrol / gas for cars, no food delivered to supermarkets, no lighting, no heating or cooling, no financial markets, and so on…
Transhumanism… might happen where humans become part human and part machine.. so part biology and part computer or non original organism components or organism adaptations. One can argue that we have already started down this path, with prosthetic limbs, enhanced eyes, artificial hips, heart and lung transplants etc etc.
Elon Musk has funded Neuralink where electrodes are planted in the brain for two way communication between your brain and computers… The author believes that the complexities of the human brain, the human immune system rejecting implants, means that such a system is decades off, however may be used sooner in humans that are paralysed or have lost the use of limbs. The author is not a fan of BCI- brain computer interfaces, and does not believe that there will be widespread adoption of devices requiring your skull to be opened up...
8) Robotics- The full system- robots with advanced AGI- replacing or augumenting humans?
Look at a youtube video of the latest Boston Dynamics robots, especially Atlas, a humanoid. Atlas is 1.5m tall, and 89kg, is starting to show the dexterity of a human, or even better than many humans. See https://youtu.be/tF4DML7FIWk, or Atlas | Boston Dynamics for a demonstration of Atlas on an obstacle course / Parkour, or Do You Love Me? - YouTube, and Boston Dynamics' amazing robots Atlas and Handle - YouTube. Or if you prefer to see fails and falls… see Robots doing Parkour is cool but watching them falls is way more fun (mashable.com) Whilst these robots are currently a little big and somewhat cumbersome, with limited capabilities, many of the things ie physical movements, a human can do can now be mechanically matched as of today.
Over time they will be made more compact and even more agile, fast and strong. If you assume that their capabilities will continue to improve and they will be made less bulky- a very reasonable assumption given the history of computers and miniaturisation, then their widespread adoption in the human workplace will depend primarily on two things:
1) their cost coming down sufficiently to compete with humans costs of doing tasks; and
2) their intelligence.. then look out, Blade Runner and Ex Machina… Incorporating AGI into an electro-mechanical system ie a robot, that can do anything a human can do better will have huge implications…
Much of these sci-fi movies will happen, the only question is when…
Photos of the Atlas Robot are below, some jumping over boxes better than many humans...
Their dog- Spot Explorer, below, can be purchased today- 2021- for $74,500. See Spot Launch - YouTube
And in 2021 (August) Elon Musk announced he was looking to build a humanoid to do dangerous, repetitive and boring tasks.. building off the work Tesla is doing in AI… Possibly just a PR stunt, but over the next year or so we will find out if there is any substance to this…
The above image is from the movie “I, Robot”, could well be the typical structure used for advanced robots, unless people prefer exact replicas of humans eg Bladerunner (Original movie)
9) Should we be worried? And AWS, Jobs etc
Maybe.
The doomsters believe that AGI will cause the extinction of the human race. James Barrat wrote the book “Our final invention” where he essentially believes that AGI will take over the world (Martin Rees is also said to be somewhat on the doomster side- Lord Martin Rees is the Astronomer Royal in the UK. He founded the Centre for the Study of Existential Risk with Huw Price and Jaan Tallinn. He is a Fellow of Trinity College and Emeritus Professor of Cosmology and Astrophysics at the University of Cambridge ) and decide that humans are a bad idea to have around as they consume resources, and therefore eliminate all humans. There are a wide range of movies exploiting such themes, such the “The Terminator”, “Bladerunner”, “I, Robot”, “Lucy” , “Transcendance” and “2001: A Space Odyssey”.. Once they “escape” they could take control of electricity, communications, fuel and water and financial markets (computers do the majority of the buying and selling on NASDAQ and the NYSE], by exploiting vulnerabilities through the internet). Imagine if stock prices dropped 80% in a day (AI overriding the circuit breakers..), and all electricity and communications shut down... The beginning of the book Life 3.0 by Max Tegmark runs through a scenario of how that may happen.. Tegmark also outlines over a dozen possible scenarios that could occur from the Polyanna to the Cassandra scenarios- see page 162 of his Life 3.0 book, and doubtless there are many more possibilities.
And we’ve learned what happens when a technologically advanced being run in to less advanced ones: Christopher Columbus versus the Tiano; Pizzaro vs the Inca; Europeans vs Native Americans etc.
An AI system can make millions of copies of itself extremely rapidly, what if it decides to do this to fulfil its goals, and controls millions of physical robots like Atlas...
Many others believe that AGI and robotics, ie Robots, will prove a huge boon / benefit for humanity, and they will become useful servants. A few movies here can have some twists but “I, Robot”, and “Bladerunner” (both the original and 2049); perhaps even “Transcendance” basically have the robots or AI being additive to humans existence, with a few twists in the plot.
Other worriers believe they could be very dangerous in a military setting, where they could make kill / destroy decisions autonomously, possibly with unintended consequences ie AWS- Autonomous Weapons Systems. Tegmark also outlines real situations where a third world war- nuclear holocaust – was only narrowly averted when computer systems alerted to a fake / computer simulated nuclear attack -which would have launched a real nuclear response and ultimately war – by the last minute intervention of humans. So newer systems may have unknown flaws in their programming / systems that then have them kill many due to incorrect programming, or alternatively another computer / AGI takes control and has them deliberately kill humans.. The next physical war (as stated, much will likely be cyber based) could have swarms of autonomous drones swooping in to exterminate thousands of people (soldiers and civilians).
Recent years have seen dramatic breakthroughs in image and speech recognition, autonomous robotics, language tasks, and game playing. The coming decades will likely see substantial progress. This promises great benefits: new scientific discoveries, cheaper and better goods and services, medical advances. Our research and collaborations have explored applications of AI across a range of global challenges, including combating climate change, pandemic response, and food security.
AI also raises near-term concerns: privacy, bias, inequality, safety and security. Using AI to breach privacy of individuals or create digital manipulations that could make anything appear real, or even manipulate elections and mindsets.
And Autonomous Weapons Systems (AWS), where weapons systems eg drones, can independently target “undesirable” or enemies and kill them as they decide… However militaries around the world are tasked with winning wars or conflicts if they occur, so they may do whatever they deem necessary to have the superior equipment… and no-one may ever know of the new machines existence until a major conflict occurs…
And jobs… Each big advance in history has brought new jobs.. muscle power was converted in to mind power… muscles used in farming were replaced by machines which then moved to mind based jobs in the city, and jobs that even in the 1980s could not have been foreseen eg web designer. The job pessimists state that this time it is different, as many of the largest employment jobs still existed 50 to 100 years ago and now many of these can now soon be replaced by machines… with driverless cars, trucks, computers you can talk to or order things, with a now decent conversation....
See, for the number employed in different USA jobs in 2019, https://stacker.com/stories/3487/most-common-jobs-america and https://stacker.com/stories/3494/most-common-jobs-america-100-years-ago
In 2019 in the USA these categories were the highest employment, and you can decide how you think this will change with the internet eg Amazon, driverless cars and trucks, much higher levels of computer intelligence, human like robotics and AGI..
Number employed 2019
Now if AGI and robots become as powerful as many think, the job displacement will be huge, with huge societal implications. If the mechanical side eg Boston Dynamics machines, becomes cheap enough, then even greater displacement will occur. Of course there could be new jobs that emerge that no-one can conceive of today, but how easily will many of the retail salespersons, or cashiers, or fast food counter workers, or customer service representatives be able to retain for these new unforeseen jobs..?
10) Summary
Like many advanced technologies that offer huge benefits for mankind, they can also offer huge threats. A few examples can include nuclear technologies, they can provide huge amounts of clean, largely pollution free, energy, yet can also provide nuclear bombs that can destroy the planet. Drones can be hugely useful for many industries such as agriculture, or Amazon deliveries etc, but when armed can be lethal weapons of war. Synthetic biology can result in many new advanced organisms to better feed or protect the planet, but in the wrong hands could result in the next horrendous biological weapon or lethal pandemic.
And the same can go for AGI. When integrated with physical robots, they could either be hugely effective weapons of war, or decide to eliminate the human race as humans are wasteful of resources, or perhaps have humans as pets… We may not see the next huge advances in AGI coming, and there is now no way we can predict the environment many years after AGI arrives.
I am optimistic for a civilisation where there will be peaceful co-existence of mankind and advanced AGI / Robotics, and the robots will hugely enhance the welfare of all mankind. Hopefully I will be right, and our children will have the most amazing life and environment in which to enjoy it!
References:
1. AI Superpowers, China, Silicon Valley and the new world order. Kai-Fu Lee (2017)
2. Genius makers, The Mavericks who brough AI to Google, Facebook and the world. Cade Metz.(2021.)
3. Our final invention. James Barrat. (2013)
4. Life 3.0. Max Tegmark. (2018)
5. The Master Algorithm. Pedro Domingos (2015)
6. Podcast, The Future of Life Institute: 11 Oct 2018 Martin Rees (Astronomer Royal)
7. Endless newspaper, podcasts and internet sources.
Comments