Project Strawberry Credit: king newswire
Introduction
In line with the exponential development of new tools to utilize artificial intelligence (AI), the company OpenAI continues to extend the boundaries of what is possible. Their latest endeavours, the Strawberry and Orion AI projects, are poised to redefine our expectations of machine intelligence and the capacity for reasoning. The associated system models represent a fundamental shift in what we can expect from AI in the future.
Definition of Reasoning
Reason is the capacity to consciously apply logic by drawing conclusions from new or existing information, with the aim of seeking truth. It is closely associated with human assets such as philosophy, science, language, mathematics and art. In the past reasoning was considered to be a distinguishing ability possessed only by humans. While reason implies a sense of rationality, reasoning is associated with the acts of thinking and cognition and involves the use of one’s intellect. The field of logic defines the ways in which humans use formal reasoning to produce logically valid arguments. Reasoning may be subdivided into concepts of logical reasoning, deductive reasoning and inductive reasoning. In some social and political settings logical and intuitive modes of reasoning may clash, while in other contexts intuition and formal reason are seen as complementary rather than adversarial. For example, in the field of mathematics, intuition is often necessary for the creative processes involved in arriving at a formal proof which is likely the most difficult task of formal reasoning. Moreover, reasoning is one of the ways whereby the process of thinking moves one idea to a related idea. For example, reasoning is the means by which a rational individual understands sensory information from its environments. This conceptualizes abstract differences such as cause and effect, truth and falsehood or ideas regarding notions of good or evil. Reasoning, as part of executive decision making, is closely identified with the ability to self-consciously change in respect to one’s goals, beliefs, attitudes and traditions, thereby fostering the capacity for freedom and self-determination. Using reason or reasoning, can also be described as the process to find the best solution to a given problem. For example, when evaluating a moral decision, ‘morality’ is, at the very least, the effort to guide one’s conduct by reason while giving equal and impartial weight to the interests of all those affected by what one decides. Psychologists and cognitive scientists attempt to study and explain how people reason, for example which cognitive and neural processes are engaged and how cultural factors affect the inferences that people make. As a result, several theories have emerged as to how reasoning affects human behaviour. However, it is too early to forecast which theory will prevail.
Difference between Intelligence and Reasoning
Reasoning, problem solving and decision-making represent different as well as overlapping aspects of human intelligence. Researchers following the cognitive psychological approach to reasoning, study the responses of a small number of participants to logical tasks such as syllogisms or formal deductions analysing a given problem. The current theories dominating psychological concepts in respect to reasoning and their relationship to intelligence are defined by mental rules and mental models. These theories were first applied to the study of deductive reasoning tasks and thereafter applied to a broader range of tasks. Human reasoning occurs at different levels of awareness. Most cognitive scientists distinguish between tacit and intentional or explicit reasoning tasks. One of the important controversies concerning reasoning and intelligence is the extent to which individuals differ in their reasoning abilities and the capacity of their neural working memory. Traditionally, IQ-tests such as the Cognitive Abilities Test (CogAT) have been used to provide a measure of cognitive development and human intelligence. However, more research is required to understand the relationship between intelligence and reasoning, especially in the context of machine vs. human reasoning.
The Promise of Strawberry and Orion
Both projects are more than just another phase providing new tech tools. They represent a fundamental shift in what we can expect from AI. As these products move from the lab to real-world applications, we are entering an era where AI becomes a collaborator, capable of tackling some of our most complex challenges. At the heart of Strawberry’s capabilities is a technique called the Self-Taught Reasoner (STaR) which moves Strawberry beyond simple pattern recognition towards the realm of genuine problem-solving. While Strawberry is impressive, it is just a first step of more to come. Orion represents the main effort in OpenAI’s drive for innovation. Sometimes also labelled GPT-5 or the next evolution in language model, it includes some of the following features that make it so special:.
Boost in Reliability: By leveraging Strawberry’s reasoning ability, Orion aims to be more dependable with fewer ‘hallucinations’ since continuous prompting for solving a problem produces increasingly stupid responses.
Enhanced Interaction: The combination of Strawberry’s problem-solving capability and Orion’s advanced architecture promises AI-interactions that are more natural, nuanced and knowledgeable.
Moreover, to gain user acceptance and to improve its capability for solving problems in the field of mathematics, OpenAI plans to integrate Strawberry into ChatGPT as early as this fall. Orion’s exact launch date is still under wraps, but industry insiders suggest that we might see the product improvements by year’s end. Once available, it is likely to revolutionize everything from customer service to scientific research.
Will AI surpass Human Reasoning?
In an Essay published by Medium on August 28,2024, the question is raised if we may have reached the point where AI-systems can outthink the majority of individual humans, thereby defining the moment in time when humans have reached the peak as an intellectual force on planet earth. Once we pass this milestone, we will steadily lose our cognitive edge until AI-systems can outthink all individual humans, even the most brilliant among us. There are many ways to measure reasoning, none-of-which is considered the gold standard, but the best known is the classic IQ test. Open AI’s so- called ‘01 product’ which combines Strawberry and Orion, has scored an average IQ of 95 when given a test it was not previously trained on. This is an impressive result because the product outperformed 37% of adults on the same reasoning tasks without support by intelligent systems. It also represents a huge increase, over Open AI’s prior model GPT-4, released just a year ago. At the current rate of technological progress, it is very likely that an AI model will be able to beat 50% of adult humans on standard IQ tests by the end of this year. From this perspective, we are likely to observe new AI-systems that exceed human reasoning. Hence, it can be anticipated that future AI-systems will be capable of outperforming all humans on reasoning tasks regardless of their individual IO. A new technology called Conversational Swarm Intelligence (CSI) enables large groups of up to 400 people to hold real-time conversational deliberations and discussions that mirror and amplify the group’s collective intelligence. In collaboration with Carnegie Mellon University, a study was conducted early 2024 whereby a group of 35 randomly selected individuals was confronted with IQ test questions, asked jointly in a real-time setting. Using a new CSI-powered online collaboration environment called Thinkscape, designed as a tool to provide AI-facilitated ‘conversational swarms’, the group’s average IQ reached an impressive score of 128. This is a strong result, but we may be just scratching the surface of how smart humans, working as groups, can be when we use AI-systems to connect them together for performing complex reasoning tasks.
Conclusion
OpenAI’s Strawberry and Orion products represent a fundamental shift in what we can expect from AI in the near future. As these models move from the lab to real-world applications, we are entering an era where AI is not just a tool, but a collaborator capable of tackling some of our most complex challenges which represents more than just an incremental advancement in technology. These products are the basis of a quantum leap in AI capabilities that could reshape education for personalized tutoring that adapts to each student’s needs, tackling complex subjects with far greater ease than was previously possible. Or they might impact software development for code generation and debugging that could dramatically speed up the development cycle and reduce cost. The future will tell if progress in machine reasoning ‘outsmarting’ humans will really signal a new age of technological progress or if we might just witness a new hype in AI-technologies.