“The first principle is that you must not fool yourself and you are the easiest person to fool.” — Richard P. Feynman
Don’t be fooled by the hype and sloppy language about Artificial Intelligence. The technology advances exponentially, but human nature remains unchanged. Living wisely requires clear thinking about what AI is, and is not.
The key ideas in one paragraph
“AI” today is all Artificial and no Intelligence. A true AI would have consciousness and self-awareness, and today’s tools have neither. Even the most sophisticated “AI” tools today are analogous to viruses – they aren’t alive and are completely dependent on the machinery built by others to work. The belief that a Singularity (a future combination of digital systems becomes independently intelligent) is coming relies on faith without evidence from physics or math. Consciousness does not spontaneously arise from parts. The digital tools we’re creating are increasingly useful, and there is great promise for improved human living by expanding our digital services. All tools humans have developed could be used for good and for evil. The existential danger of digital tools exists not because of a “Skynet” super-intelligence, but from the fallible humans creating and controlling algorithms and data.
Let’s unpack these ideas.
“AI” is all Artificial and no Intelligence
Exciting announcements about AI advances come frequently now. Siri is a voice interface which can answer your questions. AlphaGo defeats the best human Go players. The Google assistant can make appointments for you. Deep learning models can interpret X-rays better than human radiologists and make better medical diagnostics than experienced doctors, tapping into knowledge from thousands of medical research papers. Multiple companies have engineered autonomously driven vehicles (John Deere tractors have had self-driving capabilities since the 1990’s). There are digital tools which can create music, write sports stories for newspapers and web sites, and make recommendations on what you should buy. These digital tools deliver increasingly sophisticated results and iteratively improve as designers access more data and better prediction models.
All these are amazing and useful, but none of them are intelligent. They aren’t self-aware. They don’t understand anything. They are not conscious. They are clever algorithms executing on data, based on their designer’s intentions. They can “appear” to be acting intelligently, but they only reflect the intelligence of their designers.
It’s important not to confuse “analyzing a problem MUCH faster than a human” and “generating new information from existing information” with intelligence. The Deep Blue chess software[1][2][3] which defeated Gary Kasparov in 1996 could analyze millions of sequences of moves in seconds but didn’t understand chess. AlphaGo is engineered differently, using neural networks and reinforcement learning models[4][5], but is not self-aware that it is playing a game. Story-creating software creates a news story based on a handful of facts about a baseball game using templated sentence structure and a library of phrases. The Turing Test[6][7][8] is a measure of sophistication to fool another human being, but not a test for intelligence.
Language inherently requires the ability to infer meaning from a message. Inference is the process of reaching conclusions that are not explicit in what was said. The fundamental design of digital systems today is deductive by default. Anything which is predicted as an inference is a probability distribution. This is a fundamental limitation of O’s and 1’s and silicon-based circuitry. If an algorithm scans an image and “recognizes” the letter A it is because the probability distribution of the pattern is highly correlated with the letter A, which was trained into the algorithm. When you speak with Siri or Google, a complicated back-end analysis is assessing the probability of word matches by matching audio wave patterns. Siri and Google systems do not “understand” your speech.
These digital tools cannot function apart from physical computer components, memory management systems, and networks to move data packets. Remove of these human-engineered, human-supported components, and the digital tool is inactive code.
Digital tools are like a virus. The virus carries information. A virus is not alive. A virus is only active when it infects a living cell and leverages the machinery of that cell to produce more viruses.
The popular press reports on Craig Venter’s progress in creating an artificial living cell gave many people flawed perceptions of what was accomplished. DNA synthesized in a lab was pieced together to create a working bacterial genome. This was inserted to a living bacterial cell to replace the natural genome. Apart from a new instruction kit – which was mostly patterned after the natural DNA instruction pattern – everything else about this “artificial cell” was original[9][10][11]. These attempts tell researchers a great deal about how small genomes operate but they have not created artificial life.
The popular press accounts about AI are also giving citizens flawed perceptions about the nature of AI.
The “Singularity” is a Fantasy
Ray Kurzweil, a brilliant engineer and futurist, has made many successful predictions about technological advances[12][13][14]. He is probably the most famous advocate of the coming “Singularity,” when a digital intelligence emerges which rivals and then exponentially outpaces human intelligence. In his 2005 book, “The Coming Singularity” Kurzweil suggested that this would happen in 2029, based on the exponential trends of computing power. More recently, the date has been pushed out to 2040, albeit with little explanation why. [15][16][17]
We can point to two singularities which already happened. Life started. Humans appeared with unique intelligence capabilities which exceeded other primates. The origin of life, and of humans, is deeply disputed. All known life exploits DNA as a code, and we know of no codes which are not designed.[18] There is an X Prize available for anyone who can demonstrate a non-designed code[19][20][21].
People have debated the definition of life[22][23][24] and consciousness for thousands of years, without coming to agreement that would support an engineering design spec. We have multiple mental models and philosophical constructs about consciousness and intelligence.[25][26][27] Even those who agree that not all living things are conscious agree that all known consciousness is associated with a living organism. The general model that consciousness represents highly integrated information accessible by the brain is far short of an engineering model for how to create it. Kurzweil and others speak about uploading an individual’s consciousness to a digital “cloud” in the future. A working detail for how consciousness operates will be required if we’re going to engineer an artificial environment to host it.
I state confidently that a true Artificial Intelligence is not going to spontaneously arise from the kinds of computer systems we have today. This is comparable to life arising spontaneously from non-life. It doesn’t matter how many servers you have, what exponential CPU power exists, or how much data you collect. There is no understood law of physics or math which supports the prediction that self-awareness and intelligence happens at a future critical threshold.
Belief that a Singularity is coming is faith without supporting evidence.
The Exciting Potential of Digital Tools as Servants
Digital tools can be extremely helpful and powerful, even if true AI is not going to happen with our silicon-based systems. I admire and respect the optimism of techno-futurists like Peter Diamandis. I recommend his book “Bold[28].” Technology advances coupled with entrepreneurial organizations are responsible for much of the betterment of the world in every measure in the last 100 years. In 2018 more than 50% of the global population was middle or upper class economically for the first time. We produce so much food that almost no one has starved to death because of famines since 1990. Medical advances are tremendous. Literacy and schooling is available to a billion more people now than in 1980. We could list many advances like this.
Digitization has been a powerful democratizing factor in making more capabilities available to more people at a much lower cost. Digitization amplifies the technology advances from biotech, robotics, sensors, energy production/storage, and 3D printing. Digitization will be a linchpin capability in solving many of the biggest problems still facing our global population.
Yes, we have concerns about privacy of data and how data is used. Yes, parents are still figuring out how best to raise children who have access to practically any information and can spend hours in virtual reality. Yes, there are new challenges in our social fabric when even video can be convincingly manufactured to tell any story we want, and bad journalism can promote false information. Yes, the accelerating technical capabilities are transforming the workplace and eliminating whole some jobs which used to pay a living wage.
Let us be grateful for digital capabilities and aim to be wise in learning how to use it well. We’ve done this for fire, for the printing press, for electricity, and for nuclear power. I’m confident we’ll find solutions. Being a Luddite is not a constructive approach.
The Real Danger of Digital Tools
Collectivist political systems are incentivized to monitor thinking and behaviors of their citizens. The Nazis, the Soviets, and East Germans developed tremendous networks of spies and paper-based tracking systems to monitor their citizens in the 20th century. It required an enormous amount of human labor. Before the Berlin Wall came down one of every three East German citizens was an informant for the government. More recently, Cuba and Venezuela have had elaborate informant networks, though not as extensive.
The Chinese Communist Party is creating a digital, all-encompassing citizen monitoring system. Digital tools are scalable in ways that human labor is not. They won’t need a third of their population to be informants. Key components include ubiquitous video, facial recognition, mandatory tracking apps on all mobile phones, government access to all means of encryption, and systemic packet sniffing on network data flows. They intend to tie this into a “social credit” scoring system where citizens are rated based on their behavior[29][30]. A significant fraction of the government’s investment in AI is related to this monitoring capability[31].
I hope you’re as horrified about this use of digital tools as I am.
If you think “But that’s just in China,” think again. Google and Facebook have been “outed” for nefarious use of the data they’ve collected from users. The algorithms Google and Facebook (and other platforms) use both intentionally and unintentionally shape the way information is categorized and presented. Google no longer uses their famous catch-phrase “don’t be evil” in their code of conduct.[32][33][34] The US Government has been monitoring network activity of citizens under the auspices of keeping citizens safe from terrorists.[35][36][37] Amazon has large contracts to provide digital system tools to the US government.[38][39][40]
The common theme here is that a relatively small number of people have the potential to collect information, shape the way that information is accessible or presented, and could use these tools for evil purposes. Hannah Fry and others have documented how human bias shows up in the way algorithms are designed.[41]
Though we like to tell ourselves that we would never be the bad guy, history teaches us that anyone could be a participant in evil[42]. Solzhenitsyn observed “the line separating good and evil passes not through states, nor between classes, nor between political parties either — but right through every human heart — and through all human hearts.”[43] Biology and psychology teach us that everyone is limited by cognitive biases[44][45] and are therefore susceptible to manipulation by others. Digital tools greatly amplify the power of individuals to be evil.
We must guard against this current existential threat, rather than worrying about Skynet someday emerging and deciding to nuke the world. We need frameworks for transparency and a willingness to overcome our greed to make the best decisions about use of digital tools. They’re here. They’re not going to disappear. Nothing short of wise, moral people will do.
How to proceed?
- Be thoughtfully skeptical about the way AI is described and sold as a solution to problems. When you hear “AI” think “digital tool,” like the tools craftsmen and engineers use to create other tools and services. They’re useful in the right hands, but not independently intelligent.
- Tell the truth, and don’t accept people telling lies. This limits the abuses of collectivism and dictatorships.[46]
- Watch for the unintended consequences of activating many digital tools. There are always unintended consequences.
- Follow the Money and Power by asking “Who benefits?” Who benefits from the current hype about AI and the coming Singularity? Who benefits from controlling access to data and shaping how it is presented? Who benefits from controlling how the rules work (in this case, the algorithms)?
May we be wise, intelligent people consciously using our non-intelligent digital services for good.
(Glenn Brooke is the author of the soon-coming book, “Bold and Gentle:
Living Wisely in an Age of Exponential Change.” This article is adapted
from one of the chapters.)
[2] https://en.wikipedia.org/wiki/Deep_Blue_versus_Garry_Kasparov
[5] https://en.wikipedia.org/wiki/AlphaGo
[7] https://en.wikipedia.org/wiki/Turing_test
[10] https://www.technologyreview.com/s/419359/how-to-make-an-artificial-cell/
[13] https://www.wikiwand.com/en/Predictions_made_by_Ray_Kurzweil
[16] https://en.wikipedia.org/wiki/The_Singularity_Is_Near
[18] Evolution 2.0 (Perry Marshall)
[19] The Evolution 2.0 Prize
[20] https://www.herox.com/evolution2.0
[23] https://www.nasa.gov/vision/universe/starsgalaxies/life%27s_working_definition.html
[26] https://www.verywellmind.com/what-is-consciousness-2795922
[28] Bold (Peter Diamandis)
[29] Chinese “social credit” system
[30] https://en.wikipedia.org/wiki/Social_Credit_System
[31] AI Superpowers: China, Silicon Valley, and the New World Order (Kai-Fu Lee)
[33] https://gizmodo.com/google-removes-nearly-all-mentions-of-dont-be-evil-from-1826153393
[36] https://www.thenewamerican.com/usnews/constitution/item/16818-document-reveals-nsa-monitored-125-billion-phone-calls-in-one-month
[39] https://money.cnn.com/2018/04/03/investing/trump-amazon-bezos-government-contracts/index.html
[41] Hello World: Being Human in the Age of Algorithms (Hanna Fry)
[42] Man’s Search for Meaning (Victor Frankl)
[43] The Gulag Archipelago (Aleksandr Solzhenitsyn)
[45] https://en.wikipedia.org/wiki/Cognitive_bias
[46] How Do You Kill 11 Million People? (Andy Andrews)
