Will computer systems ever match or surpass human-level intelligence — and, if that’s the case, how? When the Affiliation for the Development of Synthetic Intelligence (AAAI), primarily based in Washington DC, requested its members earlier this 12 months whether or not neural networks — the present star of artificial-intelligence methods — alone might be sufficient to hit this aim, the vast majority said no. As an alternative, most mentioned, a heavy dose of an older type of AI might be wanted to get these methods as much as par: symbolic AI.
Generally known as ‘good old school AI’, symbolic AI relies on formal guidelines and an encoding of the logical relationships between ideas. Arithmetic is symbolic, for instance, as are ‘if–then’ statements and laptop coding languages akin to Python, together with circulation charts or Venn diagrams that map how, say, cats, mammals and animals are conceptually associated. Many years in the past, symbolic methods had been an early front-runner within the AI effort. Nevertheless, within the early 2010s, they had been vastly outpaced by more-flexible neural networks. These machine-learning fashions excel at learning from huge quantities of information, and underlie large language models (LLMs), in addition to chatbots akin to ChatGPT.
Now, nonetheless, the computer-science neighborhood is pushing exhausting for a greater and bolder melding of the previous and the brand new. ‘Neurosymbolic AI’ has turn into the most popular buzzword on the town. Brandon Colelough, a pc scientist on the College of Maryland in Faculty Park, has charted the meteoric rise of the idea in tutorial papers. These reveal a spike of curiosity in neurosymbolic AI that began in round 2021 and exhibits no signal of slowing down.
On supporting science journalism
Should you’re having fun with this text, think about supporting our award-winning journalism by subscribing. By buying a subscription you’re serving to to make sure the way forward for impactful tales concerning the discoveries and concepts shaping our world at the moment.
Loads of researchers are heralding the pattern as an escape from what they see as an unhealthy monopoly of neural networks in AI analysis, and count on the shift to ship smarter and extra dependable AI.
A greater melding of those two methods might result in artificial general intelligence (AGI): AI that may cause and generalize its data from one state of affairs to a different in addition to people do. It may also be helpful for high-risk functions, akin to military or medical decision-making, says Colelough. As a result of symbolic AI is clear and comprehensible to people, he says, it doesn’t endure from the ‘black box’ syndrome that may make neural networks exhausting to belief.
There are already good examples of neurosymbolic AI, together with Google DeepMind’s AlphaGeometry, a system reported final 12 months that may reliably solve maths Olympiad problems — questions aimed toward gifted secondary-school college students. However figuring out how finest to mix neural networks and symbolic AI into an all-purpose system is a formidable problem.
“You’re actually architecting this sort of two-headed beast,” says laptop scientist William Regli, additionally on the College of Maryland.
Confrontation
In 2019, laptop scientist Richard Sutton posted a brief essay entitled ‘The bitter lesson’ on his weblog (see go.nature.com/4paxykf). In it, he argued that, because the Fifties, individuals have repeatedly assumed that one of the best ways to make clever computer systems is to feed them with all of the insights that people have arrived at concerning the guidelines of the world, in fields from physics to social behaviour. The bitter capsule to swallow, wrote Sutton, is that point and time once more, symbolic strategies have been outdone by methods that use a ton of uncooked information and scaled-up computational energy to leverage ‘search and studying’. Early chess-playing computer systems, for instance, that had been skilled on human-devised methods had been outperformed by those who had been merely fed a number of sport information.
This lesson has been extensively quoted by proponents of neural networks to help the concept making these methods ever-bigger is the best path to AGI. However many researchers argue that the essay overstates its case and downplays the essential half that symbolic methods can and do play in AI. For instance, the very best chess program at the moment, Stockfish, pairs a neural community with a symbolic tree of allowable strikes.
Neural nets and symbolic algorithms each have execs and cons. Neural networks are made up of layers of nodes with weighted connections which can be adjusted throughout coaching to acknowledge patterns and be taught from information. They’re quick and creative, however they’re additionally sure to make things up and might’t reliably reply questions past the scope of their coaching information.
Symbolic methods, in the meantime, wrestle to embody ‘messy’ ideas, akin to human language, that contain huge rule databases which can be tough to construct and gradual to go looking. However their workings are clear, and they’re good at reasoning, utilizing logic to use their basic data to recent conditions.
When put to make use of in the actual world, neural networks that lack symbolic data make traditional errors: picture turbines would possibly draw individuals with six fingers on every hand as a result of they haven’t learnt the overall idea that fingers usually have 5; video turbines wrestle to make a ball bounce round a scene as a result of they haven’t learnt that gravity pulls issues downwards. Some researchers blame such errors on an absence of information or computing energy, however others say that the errors illustrate neural networks’ basic incapability to generalize data and cause logically.
Many argue that including symbolism to neural nets may be the very best — even the one — technique to inject logical reasoning into AI. The worldwide expertise agency IBM, for instance, is backing neurosymbolic strategies as a path to AGI. However others stay sceptical: Yann LeCun, one of many fathers of modern AI and chief AI scientist at tech large Meta, has mentioned that neurosymbolic approaches are “incompatible” with neural-network studying.
Sutton, who’s on the College of Alberta in Edmonton, Canada, and received the 2024 Turing prize, the equal of the Nobel prize for laptop science, holds agency to his unique argument: “The bitter lesson nonetheless applies to at the moment’s AI,” he instructed Nature. This means, he says, that “including a symbolic, extra manually crafted aspect might be a mistake.”
Gary Marcus, an AI entrepreneur, author and cognitive scientist primarily based in Vancouver, Canada, and one of the vital vocal advocates of neurosymbolic AI, tends to border this distinction of opinions as a philosophical battle that’s now being settled in his favour.
Others, akin to roboticist Leslie Kaelbling on the Massachusetts Institute of Know-how (MIT) in Cambridge, say that arguments over which view is correct are a distraction, and that folks ought to simply get on with no matter works. “I’m a magpie. I’ll do something that makes my robots higher.”
Combine and match
Past the truth that neurosymbolic AI goals to meld the advantages of neural nets with the advantages of symbolism, its definition is blurry. Neurosymbolic AI encompasses “a really giant universe,” says Marcus, “of which we’ve explored solely a tiny bit.”
There are a lot of broad approaches, which individuals have tried to categorize in varied methods. One possibility highlighted by many is using symbolic strategies to enhance neural nets. AlphaGeometry is arguably one of the most sophisticated examples of this technique: it trains a neural internet on an artificial information set of maths issues produced utilizing a symbolic laptop language, making the options simpler to examine and making certain fewer errors. It combines the 2 elegantly, says Colelough. In one other instance, ‘logic tensor networks’ present a technique to encode symbolic logic for neural networks. Statements could be assigned a fuzzy-truth worth: a quantity someplace between 1 (true) and 0 (false). This gives a framework of guidelines to assist the system cause.
One other broad method does what some would say is the reverse, utilizing neural nets to finesse symbolic algorithms. One drawback with symbolic data databases is that they’re typically so giant that they take a really very long time to go looking: the ‘tree’ of all attainable strikes in a sport of Go, for instance, comprises about 10170 positions, which is unfeasibly giant to crunch by. Neural networks could be skilled to foretell probably the most promising subset of strikes, permitting the system to chop down how a lot of the ‘tree’ it has to go looking, and thus rushing up the time it takes to choose the very best transfer. That’s what Google’s AlphaGo did when it famously outperformed a Go grandmaster.
An alternate concept is to insert symbolics into the center of an LLM’s workflow, in the identical manner as consulting a calculator would possibly assist particular person to resolve a maths puzzle. Utilizing rules-based methods throughout essential reasoning steps might help to maintain LLMs from going off-track, many argue. Initiatives together with the Program-Aided Language (PAL) mannequin, for instance, use an LLM to transform natural-language duties into Python code, use that symbolic code to resolve the issue, after which interpret that answer again into pure language with an LLM.
Jiayuan Mao, an AI researcher who has simply accomplished her PhD at MIT with Kaelbling and is on her technique to the College of Pennsylvania in Philadelphia, has had success in utilizing neurosymbolic AI to make robotic coaching extra environment friendly. Her technique is to make use of a neural community to acknowledge objects (akin to a pink rubber ball or a inexperienced glass dice) in a visible subject after which use a symbolic algorithm to cause by relational questions on these objects (akin to ‘is the rubber object behind the inexperienced object?’). A pure neural community would want 700,000 examples in its coaching information to attain 99% accuracy on this activity, she says. However by including symbolic strategies, she wants simply 10% of that quantity. “Even if you happen to use 1%, you may nonetheless get 92% accuracy, which is sort of spectacular,” she says. An analogous neurosymbolic system she created trounced a neural-network-based system at guiding a robotic that encountered unfamiliar objects whereas washing dishes or making tea.
Misplaced in translation
One of many huge challenges for symbolic AI is learn how to encode typically slippery human data inside a language of logic and guidelines. One of many earliest makes an attempt was a venture known as Cyc, began by laptop scientist Doug Lenat in 1984 and later overseen by his AI firm Cycorp, primarily based in Austin, Texas. The intent was to explicitly articulate common sense details and guidelines of thumb, akin to ‘a daughter is a baby’, ‘individuals love their kids’ and ‘seeing somebody you’re keen on makes you smile’. The venture’s language, CycL, makes use of symbols (for logical operators akin to IF, AND, OR and NOT) to precise logical relationships in order that an inference engine can simply draw conclusions, akin to ‘seeing your baby would make you smile’.
Cyc, which now holds greater than 25 million axioms, has been utilized in quite a lot of AI efforts and has impressed different tasks, akin to Google’s Information Graph, which holds greater than 500 billion details. In the present day, ‘data engineers’ use the same technique to collect human-generated details and relationships, construct specialised databases and combine them with AI.
Symbolic databases might help an AI to generalize data from one state of affairs and apply it in one other, says Kaelbling, which is a robust technique to make reasoning extra environment friendly. However there’s a trade-off in accuracy when coping with topics for which there are lots of exceptions to the ‘guidelines’ — not all individuals love their kids, for instance, and seeing one thing you’re keen on doesn’t all the time make you smile. Symbolics needs to be included solely when it’s useful to take action, she says. “Cyc was attempting to show frequent sense into math. That’s nearly certainly a nasty concept,” Kaelbling says.
In 2023, Marcus posted a paper with Lenat laying out what LLMs can be taught from Cyc. As a part of that work, the duo requested GPT-3, an early LLM of the sort that underpins ChatGPT, to put in writing CycL statements that encode the logical relationships within the sentence “Did you contact a blue object positioned within the capital of France on September twenty fifth, 2022?” The response “at first amazed the Cyc staff”, they report, as a result of it generated what appeared to be the best type of statements in the best type of language. However on nearer inspection, GPT-3 made many essential errors, they write, akin to concluding that “the factor that’s touching the blue object is the date”.
“It seems to be prefer it’s good, it seems to be prefer it ought to work, but it surely’s completely rubbish,” says Colelough. This exhibits that it’s pointless to easily ram collectively a symbolic engine and a neural internet, he says. “Then you definitely would possibly as effectively simply not have the neurosymbolic system.”
What’s wanted, Colelough says, is much more analysis on AI ‘metacognition’ — how AI displays and conducts its personal pondering. That may allow AI ‘conductors’ to supervise a extra subtle integration of the 2 paradigms, reasonably than having completely different engines merely take turns. Colelough says AlphaGeometry does this effectively, however in a restricted context. If a versatile conductor that works for any area of data might be developed, “that might be AGI for me”, Colelough says.
There’s much more work to do. Contemporary {hardware} and chip architectures may be wanted to run neurosymbolic AI effectively. In time, different kinds of AI — possibly primarily based on neural networks, symbolic AI, each or neither — would possibly turn into extra thrilling, akin to quantum AI, a fledgling subject that seeks to use the properties of the quantum world to enhance AI.
For Mao, the last word aim is to leverage neural networks’ studying talents to create guidelines, classes and paths of reasoning that people aren’t but conscious of. “The hope is that ultimately we will have methods that additionally invent their very own symbolic illustration and symbolic algorithms, in order that they’ll actually transcend what a human is aware of,” she says. That may be like a pc discovering an as-yet-unknown mathematical or bodily idea — maybe analogous to π or the property of mass — after which encoding the brand new idea to assist to increase data. “We have to examine how computer systems can educate people, not how people can educate machines.”
This text is reproduced with permission and was first published on November 25, 2025.
