{"product_id":"a-brief-history-of-intelligence-why-the-evolution-of-the-brain-holds-the-key-to-the-future-of-ai","title":"A Brief History of Intelligence: Why the Evolution of the Brain Holds the Key to the Future of AI","description":"\u003cp\u003e\u003cspan\u003e\u003cem\u003e\u003cstrong\u003e'I found this book amazing'\u003c\/strong\u003e\u003c\/em\u003e \u003cbr\u003e\u003c\/span\u003e\u003cstrong\u003eDaniel Kahneman, Winner of the Nobel Prize in Economics and bestselling author of\u003cspan\u003e \u003c\/span\u003e\u003cem\u003eThinking Fast \u0026amp; Slow\u003c\/em\u003e\u003c\/strong\u003e\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003eGABI'S REVIEW\u003c\/strong\u003e\u003c\/p\u003e\n\u003cp\u003eSince the launch of ChatGPT the rapid deployment of AI has been met with all manner of opposing claims by media hype and institutional outcry. What we know for sure is that the emerging pace of AI technology outstrips our collective ability to process potential risks, and the capacity for global deliberation to implement regulation. Ideas about regulation are not unilateral, and transparency is not governed by a central decision-making organisation. One driver of pace is the superpower AI arms race between China and the USA, resulting in what Time Magazine’s editor Sam Jacobs has called an “all gas, no brakes” scenario. So, if you are seeking an informative book to begin grappling with what AI is and is not, Max Bennett’s book\u003cem\u003e A Brief History of Intelligence\u003c\/em\u003e is a great place to start. But be prepared; this is a book on A.I., that spends most of its time pondering fish and mammals, and I would argue is all the more illuminating for doing so.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003eMax Bennett is an AI developer whose book examines the biological evolution of intelligence in a novel way. While he is not an evolutionary biologist, he painstakingly evidences the developments of intelligence in evolution and applies it to adoptive approaches in machine learning. Bennett’s central tenet is that artificial intelligence isn’t a data or compute problem, but a cognition problem. He casts the evolutionary trajectory that built the human brain through the lens of progressively expanding the aperture of flexible learning. What he is not doing in this enquiry is assigning clean functions to the brain in stages; instead, he is examining what happens when evolution adds a new component to a complex emergent network. This approach is what makes the book such an arresting read.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003eStarting with pre-brain organisms and tracking from there, Bennett outlines five emergent intelligence phases or breakthroughs as the following: steering, reinforcement, simulation, mentalising, and language. Steering endowed early organisms with goal-directed behaviour to optimise finding food and avoid becoming it. The Roomba vacuum cleaner was modelled on this first breakthrough, which is a bilateral organism steering template. Reinforcement learning was the next key phase with the adjunct appearance of neural network chemicals like dopamine and serotonin to incentivise learning from outcomes. Simulation followed: the beginning of mind models which let animals run mental what-if scenarios before committing to action. This development allowed early mammals to ‘learn’ from self. Mentalising and the development of theory of mind was the next flexible learning update, whereby an animal could internally model predictions about others and learn by imitation. This section of the book is remarkable because its revelations extend the arc of flexible learning where something vertiginous happened: an infinite regress of social modelling that underwrites primate coalition politics, including humans and allowed the final accelerant phase for intelligence “language” to operate to great effect upon the platform of all preceding phases below it.\u003c\/p\u003e\n\u003cp\u003eLanguage is so powerful an ability to enable intelligence to move beyond learning by imitating others and into the realm of transferring ideas directly.  From its declarative labelling and grammar functions, it birthed the foundation of collaborative intent and culture in humans. Grammar introduced the ordering of agreed symbols and in their re-ordering, the ability to produce different nuance and more complex meanings. The sharing of strategies and ideas through direct communication in language tethers our inner simulations to each other and turns our brains into a medium for exchange.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003eBennett’s book was completed in 2023 and so much has changed in the development of AI since, but what remains after reading this book is a thorough understanding that a Large Language Model (LLM) like ChatGPT is essentially what a calculator is to maths. It is a top-heavy intelligence, extraordinary at syntax but with no grasp of what the words refer to. Bennett’s book is singular in that he shows the evolutionary order in which embodied minds were constructed to reveal what might be important in building superintelligence when or if key steps are skipped. The current excitement around building ‘world model AI’ in robotics represents an attempt to give AI an axiomatic understanding of physics and causality, rather than the LLM statistical approximation of how for example physicists write about physics. Whether this will constitute genuine understanding or an extraordinarily sophisticated simulation of understanding is another question.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003eBennett’s book leaves you with an idea that building superintelligence is very complex; how do we ensure that superintelligence is accurate in simulating what a human would want? Empathy, for example, isn’t a value we add to intelligence when building it. It’s a structural breakthrough that evolution spent a very long-time building. You cannot produce genuine mentalising by training a model on enough descriptions of human feeling. A library of empathy is not the same as having the neural machinery that made empathy necessary in a physical world. This has consequences that extend beyond the technical. If we continue scaling systems that simulate the outputs of mentalising minds without instantiating the underlying architecture, we risk building something fluent in the vocabulary of ethics while being constitutionally incapable of its practice. It must be said the risk would not be through malice, but through the same structural absence that would make a compass useless in a world without north. It is still early days in AI and there is much to engage with in watching this fascinating tool find its place in the world.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003ePropriety is everything with this new technology: who will own it and under what terms our data is acquired and our sovereign rights protected makes AI a subject that can’t be ignored. Its great promise is the possibility of building superintelligence to avoid the pitfalls we see in primate and human behaviour, manifesting in intellectual bias and many other flawed instincts, such as our hierarchical status-seeking natures. Decentralised privacy protected AI, and open-source models may keep big tech in check; and as always, we will vote with our dollar and hopefully our informed perspectives. Intelligence is more than the ability to solve problems to achieve goals, because goals exist in philosophical territory. Bennett concludes with the most important question of all as we stand on the precipice of the sixth breakthrough of intelligence: what should be humanity’s goals? Endowed now as we are with what Bennett calls “God-like abilities of creation”, fourteen billion years in the making, he ends with this statement: “Whether we like it or not the universe has handed us the baton”.\u003cbr\u003e\u003c\/p\u003e\n\u003cp\u003e\u003cstrong\u003ePUBLISHER REVIEW\u003c\/strong\u003e\u003c\/p\u003e\n\u003cp\u003eThe entirety of the human brain’s 4-billion-year story can be summarised as the culmination of five evolutionary breakthroughs, starting from the very first brains, all the way to the modern human brains. Each breakthrough emerged from new sets of brain modifications, and equipped animals with a new suite of intellectual faculties.\u003c\/p\u003e\n\u003cp\u003eThese five breakthroughs are the organising map to this book, and they make up our itinerary for our adventure back in time. Each breakthrough also has fascinating corollaries to breakthroughs in AI. Indeed, there will be plenty of such surprises along the way. For instance: the innovation that enabled AI to beat humans in the game of Go – temporal difference reinforcement learning – was an innovation discovered by our fish ancestors over 500 million years ago. The solutions to many of the current mysteries in AI – such as ‘common sense’ – can be found in the tiny brain of a mouse. Where do emotions come from? Research suggests that they may have arisen simply as a solution to navigation in ancient worm brains. Unravelling this evolutionary story will reveal the hidden features of human intelligence and with them, just how your mind came to be.\u003c\/p\u003e","brand":"Max Bennett","offers":[{"title":"Default Title","offer_id":62708924842143,"sku":null,"price":24.99,"currency_code":"AUD","in_stock":false}],"thumbnail_url":"\/\/cdn.shopify.com\/s\/files\/1\/0394\/7236\/5727\/files\/9780008560133.jpg?v=1772681506","url":"https:\/\/lanebook.com.au\/products\/a-brief-history-of-intelligence-why-the-evolution-of-the-brain-holds-the-key-to-the-future-of-ai","provider":"The Lane Bookshop","version":"1.0","type":"link"}