Innholdsfortegnelse for EG23 i New Scientist - The Essential Guides

At the heart of current AI research are human brain-inspired computer programs called neural networks, each consisting of a vast number of connected digital neurons. We receive input through our senses, but these AIs receive input as data, such as images, audio or text. In response to an input, our brains erupt with chaotic and complex sequences of firing synapses between neurons, whereas artificial neural networks process data more methodically. A neural network can be trained to accomplish a task by passing huge amounts of data through it and then checking its output against a known expected result. For example, to create an AI that can transcribe handwriting, you could repeatedly show a neural network samples of writing and check its output against what the text is known to say.…

HUMANOID robots have just begun stepping into Amazon warehouses and Mercedes-Benz automotive factories. Now, they are being recruited for an even more ambitious effort – the creation of artificial general intelligence with capabilities comparable to those of humans. In early 2024, US computing firm Nvidia, which has become one of the world’s most valuable companies through its AI chip sales, announced several hardware and software products to boost humanoid robot training. The centrepiece is a “moonshot” initiative, called Project GR00T, focused on developing an AI model to help humanoid robots learn more efficiently through language instructions, watching human demonstrations and practising various actions with the help of human teleoperators. Robotic interactions with the physical world could, in turn, improve the AI model through an “embodied intelligence” strategy – a cornerstone…

BROADLY speaking, forecasters are predicting that generative AI will boost economic productivity and growth in advanced economies. A report released by Goldman Sachs in March 2023 predicted that generative AI could, within a decade, raise annual global GDP by 7 per cent, which translates to a roughly $7 trillion increase. The report suggested generative AI’s arrival “raises the possibility of a labor productivity boom like those that followed the emergence of earlier general-purpose technologies like the electric motor and personal computer”. The idea is that AI will make millions of “knowledge workers”, like scientists, editors, lawyers and doctors, more productive within a few years. But the truth is that these things are hard to predict and assess, especially as the output of such workers is notoriously difficult to measure. One…

WHILE there are no plans for AI judges yet, this is a question that the UK government is already wrestling with as it considers the potential uses of AI in the English court system. Despite this, lawyers and computer scientists are warning that current systems can’t handle the ambiguity and nuance often required in legal situations. For example, Vinija Jain, formerly at Stanford University in California and now working at Amazon, and her colleagues assessed how large language models (LLMs), of the type that powers ChatGPT, would judge people carrying out ethically ambiguous actions driven by their circumstances, such as stealing food to feed starving family members. While people may be split on whether or not poverty or need was mitigation in these scenarios, the AIs had none of the…

ACCORDING to David Chalmers, a philosopher at New York University, we have no solid reason to rule out some form of inner experience emerging in silicon transistors. “No one knows exactly what capacities consciousness necessarily goes along with,” he said at the Science of Consciousness Conference in Taormina, Italy, in May 2023. But sentient machines that exhibit consciousness still aren’t a settled matter – nor is an agreed definition of how we will know we are there, if ever. What we can say is that unnervingly intelligent behaviour has already emerged in these AIs. The large language models (LLMs) that underpin the new breed of chatbots can write computer code and can seem to reason: they can tell you a joke and then explain why it is funny, for instance.…

DATA-DRIVEN science began in the 16th century with Danish astronomer Tycho Brahe’s meticulous observations of the motions of planets and stars. Poring over Brahe’s notebooks, Johannes Kepler spotted underlying patterns in the data to come up with three simple laws that describe planetary motion – and confirmed the sun’s place at the centre of the universe. Kepler’s strategy was trial and error. He painstakingly fit the solar system. Eventually, he uncovered a precise mathematical harmony between the paths of planets and the time it took them to orbit the sun. It was a triumph that put mathematical equations at the heart of our understanding of the universe, where they have remained. “The great book of nature is written in mathematical language,” wrote Kepler’s contemporary and fellow heliocentrist Galileo Galilei. These…

If you really want to get to grips with how the era of AI will pan out for us, you can do little better than asking Scott Aaronson and Boaz Barak, computer scientists at the University of Texas at Austin and Harvard University, respectively. They have already gone to the trouble of boiling down the lengthy discourse on the matter to a handful of outcomes – and given each a whimsical name. These five worlds of AI, they say, can help people in the field talk meaningfully to each other about end goals and regulation. “We took a humorous tone, but we are serious about trying to ground the debate,” says Barak. THE question of how AI will shape our future begins with a simple question: will AI’s potential fizzle…

In 2003, Nick Bostrom, a philosopher at the University of Oxford, posed a thought experiment. Imagine a superintelligent AI has been set the goal of producing as many paper clips as possible. Bostrom suggested it could quickly decide that killing all humans was pivotal to its mission, both because they might switch it off and because they are full of atoms that could be converted into more paper clips. The scenario is absurd, of course, but illustrates a troubling problem: AIs don’t “think” like us and, if we aren’t extremely careful about spelling out what we want them to do, they can behave in unexpected and harmful ways. “The system will optimise what you actually specified, but not what you intended,” says Brian Christian, author of The Alignment Problem and…

If AI is something that outperforms humans by definition, it seems logical to trust it to identify people who should be stopped and searched via facial recognition, say, or to make judgements on which offenders should get probation. If it is solely about algorithms, it becomes a lot easier to sweep aside questions of bias and injustice as mere technical issues. Far from being something abstract and objective, argues Kate Crawford, AI is both material and intrinsically linked to power structures. The way it is made involves extracting resources from people and the planet, and the way it is used reflects the beliefs and biases of those who wield it. Only when we come to terms with this, she says, will we be able to chart a just and sustainable…

OpenAI’s chatbot ChatGPT is powered by a type of neural network called a transformer, which was invented by Google in 2017. A transformer improves a standard neural network by attempting to pay attention to the most important parts of a data input while it is assessing the rest methodically. In a small way, it creates context. AIs like ChatGPT are impressive, but their results are largely down to scale and vast sets of training data, says Samuel Bowman at New York University. “There’s no big, deep innovation, there’s not one big idea behind transformers. It’s just a particular way of fitting things together that happens to match really nicely with hardware and reliability in order to run really large-scale experiments.” There are other approaches that could bring slight improvements, but…

AT THE heart of the boom in AI chatbots are large language models (LLMs) – statistical models, or mathematical representations of data, designed to make predictions about which words are likely to appear together. LLMs are created by feeding huge amounts of text to a class of algorithms called deep neural networks, which are loosely inspired by the brain. The models learn complex linguistic patterns by playing a simple game: the algorithm takes a passage of text, randomly masks out some words and then tries to fill in the gaps. They are, in short, trained to predict the next word. And by repeating the process over and over, they can build up sophisticated models of how language works, says Mirella Lapata at the University of Edinburgh, UK. Recent breakthroughs are…

Scaling up current technology can create more advanced LLMs, but computing power isn’t the only factor driving generative AI. That’s because artificial intelligence models behind popular chatbots are developing faster than Moore’s law, a measure of how quickly computer hardware performance increases. This suggests that the large language models (LLMs) behind AI systems are becoming smarter at doing more with less. The revelation was made in March 2024 by Tamay Besiroglu at the Massachusetts Institute of Technology. There are two ways to improve AI model performance, he says. One is to scale up the size of an LLM, which, in turn, requires a commensurate increase in computing power. But due to the generative AI revolution, there are global supply shortages in the graphics processing unit computer chips used to power…

Neuromorphic computers, which mimic the operation of the human brain, could run more sophisticated AI models than is possible on conventional computers. Expectations for neuromorphic computers are high because they are inherently different to traditional machines. While a regular computer uses its processor to carry out operations and stores data in separate memory, a neuromorphic device uses artificial neurons to both store and compute, just as our brains do. This removes the need to shuttle data back and forth between components, which can be a bottleneck for current computers. This architecture could be less energy intensive and could also unlock new ways to train and run AI models that use chains of neurons, as real brains process information, rather than mechanically passing an input through each and every layer of…

ONE company that has consistently shown a knack for solving real-world problems is Google’s DeepMind. One of its most surprising achievements has been in protein folding. Determining the crumpled shapes of proteins based on their sequences of constituent amino acids had been a persistent problem for decades, with researchers often taking years to solve a single one. DeepMind transformed biology in 2022 by announcing that it had predicted the structure of nearly all proteins known to science in just 18 months. The team had trained its AI AlphaFold on data from known protein shapes and it learned to predict what the unsolved proteins would look like. The data is already helping researchers make advances in everything from new treatments for malaria to the creation of enzymes that can break down…

You have described your work as making space probes more curious. What do you mean by that? The smarts we’ve been able to put on spacecraft and rovers thus far have been to recognise things we understand–to target specific types of rocks with a laser or to search for dust devils in a sequence of images, for example. In the future, when we travel to the complete unknown, we will need to go beyond this. We’ll need to look for patterns in data. For example, on Earth, we might look at overhead imagery and cluster it based on colour, texture, ruggedness and linear features. Based on these features, we might naturally discriminate between lakes, rivers, mountains, forests. But on another planet or moon, these might correspond to different types of…

IN MAY 2023, computer scientist Geoffrey Hinton, known as the “godfather of AI”, quit his role at Google to warn of the “existential threat” posed by artificial intelligence. The Center for AI Safety followed up with an open letter, signed by Hinton and hundreds of others, warning that advanced AI could destroy humanity. “Mitigating the risk of extinction from AI should be a global priority,” read the statement. This sudden surge of concern seems to have been motivated by the rapid advance of AI-powered chatbots like ChatGPT and the race to build more powerful systems. The fear is that the tech industry is recklessly accelerating the escalation of AI’s capabilities. All of which sounds scary. But the warnings are also suspiciously vague. When you scrutinise the scenarios typically put forward…

THIS moment for artificial intelligence is unlike any that has come before. Powerful language-based AIs have lurched forward in ability and can now produce reams of plausible prose that often can’t be distinguished from text written by humans. They can answer tricky technical questions, such as those posed to lawyers and computer programmers. They can even help better train other AIs. At the centre of it all are large language models and other types of generative AI that can create text and images in response to human prompts. Start-ups backed by the world’s most powerful tech firms have been accelerating the deployment of these generative AIs since 2022 – giving millions of people access to convincing but often inaccurate chatbots, while flooding the internet with AI-generated writing and imagery in…

INSTEAD of building ever-larger AIs based on the same blueprint, some researchers want to turn back to nature. One path is to double down on efforts to copy the brain, better replicating the intricacies of real brain cells and the ways their activity is choreographed. But the brain is the most complex object in the known universe and it is far from clear how much of its complexity we need to replicate to reproduce its capabilities. That’s why some believe that more abstract ideas about how intelligence works can provide shortcuts. Their claim is that to really accelerate the progress of AI towards something that we can justifiably say thinks like a human, we need to emulate not the brain, but the mind. Whether the mind and the brain can…

ARTIFICIAL intelligence is good at language. Today, our email services can complete sentences for us, our browsers automatically translate web pages and voice assistants decode our commands. The tech behind ChatGPT can write coherent sentences. Decoding animal communication is just a logical next step, says Michael Bronstein at Imperial College London. “I think it’s the right time, with the right data and with the right expertise, to possibly solve this problem.” The latest AIs learn linguistic patterns from huge amounts of human-supplied language data, without any clue how our languages actually work. Essentially, they create a vast, multidimensional “cloud” of words clustered according to how we use them, and this lets them decode new snippets of text. In 2018, researchers at Facebook realised that if you twist the clouds for…