On Whether Generative AI And Large Language Models Are Better At Inductive Reasoning Or Deductive Reasoning And What This Foretells About The Future Of AI

These programs, such as Constructive Solid Geometry (CSG), Computer-Aided Design (CAD), and Scalable Vector Graphics (SVG), provide a clear and interpretable representation of visual content. Moreover, LLMs have been applied to various programming tasks, such as code retrieval, automated testing, and generation; however, understanding symbolic graphics programs is largely different, as their semantic meaning is often defined visually. Existing benchmarks for LLMs focus on non-graphics program understanding, while vision-language models are evaluated using multimodal datasets for tasks like image captioning and visual question answering. DeepMind’s program, named AlphaGeometry, combines a language model with a type of AI called a symbolic engine, which uses symbols and logical rules to make deductions.

Although word units can be discovered from character strings using unsupervised learning (Goldwater et al., 2009; Mochihashi et al., 2009), such approaches have been extended to consider speech input as an observation. With recent advancements in deep RL, flexibly interpreting the meaning of signals issued in communication using the representational capability of deep learning has been possible symbolic ai Buşoniu et al. (2010). If this is true, then predictive learning of these collections of sequences indirectly models the world experiences we obtain through our human sensorimotor systems. In other words, the latent structure embedded in large-scale language corpora as distributional semantics, which can be learned through language modeling, represents the latent structure of the world.

To further test the hypothesis, it is essential to develop computational simulations based on CPC principles and compare the results with human data to validate the model’s predictions. For the first time, CPC offers a framework that can theoretically and comprehensively capture the entire picture of a symbolic emergence system. By capturing the dynamics of both cognition and society, CPC can holistically explain the dynamics by which language emerges in human society.

• Human infants learn symbolic communication, including language, through interaction with their environment during their developmental stages.
• Dr. Hinton, often called the godfather of AI, warns that as AI systems begin to exceed human intellectual abilities, we face unprecedented challenges in controlling them.
• The tested LLMs fared much worse, though, when the Apple researchers modified the GSM-Symbolic benchmark by adding “seemingly relevant but ultimately inconsequential statements” to the questions.
• Subtleties in the algorithms, data structures, ANN, and data training could impact the inductive reasoning proclivities.
• The internal representation learning process or categorization begins before learning linguistic signs, such as words (Quinn et al., 2001; Bornstein and Arterberry, 2010; Junge et al., 2018).

Yet, despite these warnings, venture capitalists (VCs) have been pouring billions into LLM startups like lemmings heading off a cliff. The allure of LLMs, driven by the fear of missing out on the next AI gold rush, has led to a frenzy of investment. VCs are chasing the hype without fully appreciating the fact that LLMs may have already peaked.

Generating 100 million synthetic data examples

A software component known as the inference engine then applied that knowledge to solve new problems within the subject domain, with a trail of evidence providing a form of explanation. Five years later, came the first published use of the phrase “artificial intelligence” in a proposal for the Dartmouth Summer Research Project on Artificial Intelligence. As a Tech enthusiast, he delves into the practical applications of AI with a focus on understanding the impact of AI technologies and their real-world implications. This case study exemplifies how Neuro-Symbolic AI can transform customer service by leveraging the strengths of both symbolic and neural approaches.

Remember for example when I mentioned that a youngster using deductive reasoning about the relationship between clouds and temperatures might have formulated a hypothesis or premise by first using inductive reasoning? If so, it is difficult to say which reasoning approach was doing the hard work in solving the problem since both approaches were potentially being undertaken at the same time. Their experiment consisted of coming up with tasks for generative AI to solve, along with prompting generative AI to do the solution process by each of the two respective reasoning processes. After doing so, the solutions provided by AI could be compared to ascertain whether inductive reasoning (as performed by the AI) or deductive reasoning (as performed by the AI) did a better job of solving the presented problems. For those of you familiar with the history of AI, there was a period when the symbolic approach was considered top of the heap. This was the era of expert systems (ES), rules-based systems (RBS), and often known as knowledge-based management systems (KBMS).

However, the emergence of phonological systems is not discussed, although word discovery is mentioned in relation to speech signals in Section 3.2, from the viewpoint of PC by a single agent. Computational models for the self-organization of speech codes in multi-agent systems have also been studied for more than a decade (Oudeyer, 2005). In particular, the work by Moulin-Frier et al. (2015) proposed a Bayesian framework for speech communication and the emergence of a phonological system, termed COSMO (Communicating about Objects using Sensory–Motor Operations). Integrating this concept into the CPC framework may provide a possible path for creating a more general computational model for SESs.

Collective predictive coding hypothesis: symbol emergence as decentralized Bayesian inference

In February, Demis Hassabis, the CEO of Google‘s DeepMind AI research lab, warned that throwing increasing amounts of compute at the types of AI algorithms in wide use today could lead to diminishing returns. Getting to the “next level” of AI, as it were, Hassabis said, will instead require fundamental research breakthroughs that yield viable alternatives to today’s entrenched approaches. John Stuart Mill championed ethical considerations long before the digital age, emphasizing fairness … In the medical field, neuro-symbolic AI could combine clinical guidelines with individual patient data to suggest more personalized treatment options.

However, these methods often introduce biases or require extensive optimization and hyperparameter tuning, resulting in long training times and reduced applicability to complex tasks. Moreover, these approaches generally need stronger guarantees of the accuracy of their approximations, raising concerns about their outcomes’ reliability. Neuro-Symbolic Artificial Intelligence (AI) represents an exciting frontier in the field. It merges the robustness of symbolic reasoning with the adaptive learning capabilities of neural networks. This integration aims to harness the strong points of symbolic and neural approaches to create more versatile and reliable AI systems.

“Our results demonstrate the effectiveness of the proposed agent symbolic learning framework to optimize and design prompts and tools, as well as update the overall agent pipeline by learning from training data,” the researchers write. To address these limitations, researchers propose the “agent symbolic learning” framework, inspired by the learning procedure used for training neural networks. AI agents are showing impressive capabilities in tackling real-world tasks by combining large language models (LLM) with tools and multi-step pipelines. LLM agents might one day be able to perform complex tasks autonomously with little or no human oversight. For a while now, companies like OpenAI and Google have been touting advanced “reasoning” capabilities as the next big step in their latest artificial intelligence models. Now, though, a new study from six Apple engineers shows that the mathematical “reasoning” displayed by advanced large language models can be extremely brittle and unreliable in the face of seemingly trivial changes to common benchmark problems.

Researchers are now seeking ways to transition from this engineering-centric approach to a more data-centric learning paradigm for language agent development. ChatGPT is a large language model (LLM) constructed using either GPT-3.5 or GPT-4, built upon Google’s transformer architecture. It is optimized for conversational use through a blend of supervised and reinforcement learning methods (Liu et al., 2023). However, due to the statistical nature of LLMs, they face significant limitations when handling structured tasks that rely on symbolic reasoning (Binz and Schulz, 2023; Chen X. et al., 2023; Hammond and Leake, 2023; Titus, 2023). For example, ChatGPT 4 (with a Wolfram plug-in that allows to solve math problems symbolically) when asked (November 2023) “How many times does the digit 9 appear from 1 to 100?

These advancements make AlphaGeometry 2 a powerful tool for solving intricate geometric problems, setting a new standard in the field. Transformer networks have come to prominence through models such as GPT4 (Generative Pre-trained Transformer 4) and its text-based version, ChatGPT. These large-language models (LLMs) have been trained on enormous datasets, drawn from the Internet. Human feedback improves their performance further still through so-called reinforcement learning. “The idea that these language models just store a whole bunch of text, that they train on them and pastiche them together — that idea is nonsense,” he said.

In this post, I discuss how the current hurdles of Generative AI systems could be (have been?) mitigated with the help of the good old symbolic reasoning. The ethical challenges that have plagued LLMs—such as bias, misinformation, and their potential for misuse—are also being tackled head-on in the next wave of AI research. The future of AI will depend on how well we can align these systems with human values and ensure they produce accurate, fair, and unbiased results. Solving these issues will be critical for the widespread adoption of AI in high-stakes industries like healthcare, law, and education. The fact that LLMs are hitting their limits is just a natural part of how exponential technologies evolve. Every major technological breakthrough follows a predictable pattern, often called the S-curve of innovation.

While individual cognition, development, learning, and behavior undoubtedly underpin language learning and its use, the language cultivated within society and the dynamics that support it extend beyond individual cognition. To overcome these challenges, we propose the CPC hypothesis, which radically extends the concept of PC (Hohwy, 2013; Ciria et al., 2021). This hypothesis expands PC from a single brain to a group of brains, suggesting a multi-agent system. It posits that the symbol system emerges as a result of CPC conducted collaboratively by agents in a decentralized manner. In this framework, the emergent symbol system, namely, language, is viewed as a kind of subject, akin to a brain in PC. Within the CPC hypothesis, language is considered a form of collective intelligence, implying that LLMs are directly modeling this collective intelligence.

This fusion gives users a clearer insight into the AI system’s reasoning, building trust and simplifying further system improvements. Neuro-symbolic AI combines today’s neural networks, which excel at recognizing patterns in images like balloons or cakes at a birthday party, with rule-based reasoning. This blend not only enables AI to categorize photos based on visual cues but also to organize them by contextual details such as the event date or the family members present. Such an integration promises a more nuanced and user-centric approach to managing digital memories, leveraging the strengths of both technologies for superior functionality. AI systems often struggle with complex problems in geometry and mathematics due to a lack of reasoning skills and training data. AlphaGeometry’s system combines the predictive power of a neural language model with a rule-bound deduction engine, which work in tandem to find solutions.

Prior to joining Bosch, he earned a PhD in Computer Science from WSU, where he worked at the Kno.e.sis Center applying semantic technologies to represent and manage sensor data on the Web. Leaders must develop a clear understanding of the strengths and limitations of their AI toolkit and, if they’re going to add lasting value, make a commitment to building systems that are transparent, explainable and—most importantly—trustworthy. Decision intelligence is the discipline of making better decisions with the help of machines, and it’s once again on the rise in the enterprise world—not least because it advocates for a hybrid approach. Relying solely on LLMs for decision-making remains an incomplete and risky approach, especially in regulated domains where accountability and transparency are paramount.

Alessandro’s primary interest is to investigate how semantic resources can be integrated with data-driven algorithms, and help humans and machines make sense of the physical and digital worlds. Alessandro holds a PhD in Cognitive Science from the University of Trento (Italy). Knowledge graph embedding (KGE) is a machine learning task of learning a latent, continuous vector space representation of the nodes and edges in a knowledge graph (KG) that preserves their semantic meaning. This learned embedding representation of prior knowledge can be applied to and benefit a wide variety of neuro-symbolic AI tasks. One task of particular importance is known as knowledge completion (i.e., link prediction) which has the objective of inferring new knowledge, or facts, based on existing KG structure and semantics.

This form of AI, akin to human “System 2” thinking, is characterized by deliberate, logical reasoning, making it indispensable in environments where transparency and structured decision-making are paramount. Use cases include expert systems such as medical diagnosis and natural language processing that understand and generate human language. Early deep learning systems focused on simple classification tasks like recognizing cats in videos or categorizing animals in images. Now, researchers are looking at how to integrate these two approaches at a more granular level for discovering proteins, discerning business processes and reasoning. Furthermore, the emergence of speech codes, such as phonological systems, is an important topic in the study of SESs. This study focuses on the emergence of the semantic aspects of symbol systems.

Artificial Intelligence

The topic has garnered much interest over the last several years, including at Bosch where researchers across the globe are focusing on these methods. In this short article, we will attempt to describe and discuss the value of neuro-symbolic AI with particular emphasis on its application for scene understanding. In particular, we will highlight two applications of the technology for autonomous driving and traffic monitoring.

In a computational experiment conducted in a simulated environment, a group of agents created ways to identify each other using vocabulary-related spatial concepts (Steels, 1995). Steels and Belpaeme (2005) proposed a variety of models to examine mechanisms through which a population of autonomous agents could arrive at a repertoire of perceptually grounded categories. In real-world environments, Steels et al. conducted the “Talking Heads” experiment, where each agent grounded a lexicon to a concept based on visual information to develop a method of communication among agents (Steels, 2015). These experiments showed that language games allowed agents to share lexicons and meanings of simple objects, such as red circles and blue rectangles. Mobile robots (e.g., AIBO), which have numerous modalities and behavioral capabilities, were used in experiments to learn words and meanings of simple objects and spatial concepts (Steels and Kaplan, 2000; Steels and Loetzsch, 2008). Spranger et al. studied the evolution of grounded spatial languages within a language-game framework (Spranger, 2011; 2015).

The FEP proposed by Friston was a generalization of PC (Friston, 2019), which is a general and powerful concept. Based on the variational inference perspective, multi-modal categorization (i.e., internal representation learning) and the accompanying optimization criteria are discussed. Bilateral feedback between higher and lower layers is called the micro–macro loop (or effect) (Figure 2). A pattern (or order) in a higher layer is organized in a bottom-up manner through interactions in the lower layer, and the organized pattern imposes top-down constraints on the interactions of the lower layer. This bilateral feedback provides functionality to the system, and the loop is a feature of a complex system with an emergent property used to obtain a function that is not originally discovered by the agents in the lower layer. Taniguchi et al. argued that symbolic communication emerges as a function of the micro–macro loop in complex systems.

DeepMind says this system demonstrates AI’s ability to reason and discover new mathematical knowledge. Its performance matches the smartest high school mathematicians and is much stronger than the previous state-of-the-art system. Beyond personal efficiency, anyone who is starting with the question of how they should deploy more generative AI is starting from the wrong premise; LLMs are but one piece in a much bigger and more interesting puzzle. AiThority.com covers AI technology news, editorial insights and digital marketing trends from around the globe. Updates on modern marketing tech adoption, AI interviews, tech articles and events.

Humans communicate using complex languages that involve numerous characteristics such as syntax, semantics, and pragmatics. Notably, the meaning of a sign can change through long-term interactions with the environment and other agents, depending on the context. The adaptability ChatGPT and emergent properties of symbol systems are crucial in human symbolic communication in relation to the principles of semiotics as outlined by Peirce (Chandler, 2002). Peirce emphasizes the evolutionary nature of symbols in relation to human development.

The findings highlight that these models rely more on pattern recognition than genuine logical reasoning, a vulnerability that becomes more apparent with the introduction of a new benchmark called GSM-Symbolic. Mathematical reasoning and learning meet intricate demands, setting crucial benchmarks in the quest to develop artificial general intelligence (AGI) capable of matching or surpassing human intellect. A major challenge involves how to best connect them into one cohesive mechanization.

AlphaProof and AlphaGeometry 2 have showcased impressive advancements in AI’s mathematical problem-solving abilities. However, these systems still rely on human experts to translate mathematical problems into formal language for processing. As a visionary entrepreneur and engineer, Asif is committed to harnessing the potential of Artificial Intelligence for social good. The platform boasts of over 2 million monthly views, illustrating its popularity among audiences. There are many positive and exciting potential applications for AI, but a look at the history shows that machine learning is not the only tool.

The upshot is that generative AI is likely better at inductive reasoning and that it might take some added effort or contortions to do deductive reasoning. When using generative AI, you can tell the AI via a prompt to make use of deductive reasoning. Similarly, you can enter a prompt telling the AI to use inductive reasoning. It could be that the actual internal process is nothing like the logical reasoning we think it is.

Computational models can be developed to enable AIs and robots to perform symbol emergence in a variety of tasks to test the feasibility of the CPC hypothesis in a constructive manner. Psychological experiments can also be conducted to determine whether humans actually perform the learning processes assumed in the CPC hypothesis. Particularly, Hagiwara et al. (2019) assumed that agents decide whether to accept or reject another agent’s utterance using a certain probability calculated based on their individual beliefs.

The researchers opted to explore whether inductive reasoning or deductive reasoning is the greater challenge for such AI. Their research underscores the importance of continuous innovation and refinement in developing AI models for music generation. By delving into the nuances of symbolic music representation and training methodologies, they strive to push the boundaries of what is achievable in AI-generated music. Through ongoing exploration of novel tokenization techniques, such as ABC notation, and meticulous optimization of training processes, they aim to unlock new levels of structural coherence and expressive richness in AI-generated compositions. Ultimately, their efforts not only contribute to advancing the field of AI in music but also hold the promise of enhancing human-AI collaboration in creative endeavors, ushering in a new era of musical exploration and innovation. Advocates for this approach highlight ABC notation’s inherent readability and structural coherence, underscoring its potential to enhance the fidelity of musical representations.

To understand the limitations of generative AI, it’s essential to look back at the history of AI. In the 1980s and 1990s, we had an era of ChatGPT App that, despite its limitations, was transparent and grounded in explicit rules and logic. It powered expert systems that generated a clear chain of reasoning for their outputs and was particularly adept at tasks requiring structured problem-solving. You can foun additiona information about ai customer service and artificial intelligence and NLP. Existing studies demonstrated that PGM-based approach could achieve word discovery and lexical acquisition from continuous perceptual sensory information.

• It will undoubtedly become crucial for lawyers to master AI tools, but these tools are most effective when wielded by those with uniquely human strengths.
• The open-sourcing of code and prompts aims to accelerate progress in this field, potentially revolutionizing language agent development and applications.
• This creates systems that can learn from real-world data and apply logical reasoning simultaneously.
• The competition not only showcases young talent, but has emerged as a testing ground for advanced AI systems in math and reasoning.
• People are taught that they must come up with justifications and explanations for their behavior.

The perspectives offered by the CPC hypothesis give us new speculative thoughts on this question. The demand for systems that not only deliver answers but also explain their reasoning transparently and reliably is becoming critical, especially in contexts where AI is used for crucial decision-making. Organizations bear a responsibility to explore and utilize AI responsibly, and the emphasis on trust is growing as AI leaders seek new ways of leveraging LLMs safely.

It would be immensely interesting to see the experimental results if various prompting strategies were used. Another worthy point to bring up is that I said earlier that either or both of those reasoning methods might not necessarily produce the right conclusion. The act of having and using a bona fide method does not guarantee a correct response. That being said, if you have in fact managed to assemble Lego bricks into a human-like reasoning capacity, please let me know. Live Science is part of Future US Inc, an international media group and leading digital publisher. In their simplest form, AI tokens mimic tokens from free-standing video games in arcades.

It then employs logical reasoning to produce answers with a causal rationale. When utilized carefully, LLMs massively augment the efficiency of experts, but humans must remain “to the right” of each prediction. LLMs are amazing word prediction machines but lack the capability to assess problems logically and contextually. They also don’t provide a chain of reasoning capable of proving that responses are accurate and logical.

Now, AI is evolving to emulate this duality, potentially reshaping legal work. “Processing time evidence for a default-interventionist model of probability judgments,” in Proceedings of the Annual Meeting of the Cognitive Science Society (Amsterdam), 1792–1797. “Control regularization for reduced variance reinforcement learning,” in International Conference on Machine Learning (Long Beach, CA), 1141–1150. As each Olympiad features six problems, only two of which are typically focused on geometry, AlphaGeometry can only be applied to one-third of the problems at a given Olympiad. Nevertheless, its geometry capability alone makes it the first AI model in the world capable of passing the bronze medal threshold of the IMO in 2000 and 2015.