In the past few hours, many online newspapers have literally competed to describe the potential threat represented by the OpenAI Q* project.
The development of the Q* project became the focal point of the scandal, leading the previous board of directors of OpenAI to oust the historic CEO Sam Altman, who was subsequently reinstated following pressure from Microsoft and its CEO Satya Nadella.
What is OpenAI's Q-Star?
We observed that the company's intent in the medium to long term was to develop Artificial General Intelligence (AGI). AGI is a type of artificial intelligence that can understand, learn, and apply knowledge in a similar way to a human. Unlike many specialized AI systems that perform specific tasks, an AGI should have the ability to tackle a wide range of complex tasks autonomously without being limited to a single domain of expertise.
The goal of an AGI is to replicate the breadth and flexibility of human intelligence, demonstrating the ability to adapt to new situations, learn from different experiences, and apply the knowledge developed across multiple contexts.
The idea of AGI raises ethical and security questions, as such advanced intelligence could have significant impacts on society as a whole.
OpenAI will never be the same again
Among the various reasons that would have led to the dismissal of Sam Altman, there would be a heavy ideological conflict between the CEO and the OpenAI board regarding the speed of technological development within the company.
What is certain is that OpenAI was born as a non-profit organization (and still is today): the model was chosen with the intention of giving priority to the mission of pursuing the public good over the needs of 'making money.' However, something also confirmed by the 'week of madness' that OpenAI has just experienced, ethics has clearly come into conflict with the economic interests of investors, with the prevalence of financial interests.
This is why OpenAI will never be the same again; it already isn't. It is now a reality clearly oriented towards satisfying the appetites of investors, and it's hard to think that it could be anything different.
Our strong skepticism about Q-Star
There are many 'detailed analyses' that describe Q* as a sort of 'Death Star' poised to annihilate us all. While the issue of safety and ethics in the development of artificial intelligence is undoubtedly central, it is essential to keep our feet firmly on the ground.
Q* is presented as an AGI capable of solving some mathematical problems in a deterministic way. Today, it would be able to perform calculations like an elementary school child, but in the future, its skills could grow dramatically.
As we have repeatedly highlighted, current generative models are based on a stochastic approach. Their ability to generate output is not completely determined by a fixed set of parameters or initial conditions. In contrast, these models introduce probabilistic elements into the generation process, making the output variable even when the initial conditions are the same. This is a winning scheme because it introduces diversity and variability in the results, making the model more flexible and capable of producing richer and more realistic outputs.
As we know, stochastic generative models are based on machine learning techniques such as generative neural networks (GANs), recurrent neural networks (RNNs), or probabilistic language models. These models define a probability distribution on the training data and use this distribution to generate new data that resembles the training data.
Differences between deterministic and stochastic approaches
The deterministic approach relies on fixed rules and initial conditions to produce repeatable and predictable results. In a deterministic context, the outcome of a process is completely determined by the initial configuration and the rules that guide the process. This scheme is used when it is essential to obtain consistent and easily interpretable results, especially in contexts where clarity and reproducibility are priorities.
In contrast, the stochastic approach is based on the richness of the data and exploits the element of randomness or probability to introduce variability in the results. Using probability distributions and random sampling techniques, stochastic models can generate a wide range of outcomes, making it possible to produce more varied and realistic outputs. This is especially useful when it comes to dealing with the complexity and diversity of the real world. In creative contexts, such as generating text, images, or sounds, the stochastic approach can therefore produce creative and original results.
By using the stochastic approach in modern artificial intelligence, nothing is discarded. The deviation from a deterministic model becomes a peculiarity. The data, of whatever kind, thus automatically increases in value. The more data you pass to the algorithm, the more information you can derive, and the more valuable content you can automatically produce.
Data culture plays a central role
An AI can certainly be designed to perform computations deterministically rather than stochastically. If the initial conditions and rules are known, the result of the processing will always be the same. In the context of AI, a model or algorithm that operates deterministically will consistently produce the same output for a specific input, regardless of how many times it is run.
In the vast majority of real-world situations, the environment in which an AI finds itself operating is complex and dynamic. Unexpected variables, complex interactions, and rapid changes can make it difficult to ensure fully deterministic behavior.
Trending News
• Amazon launches AI ready initiative to train millions in AI skills
• Google's GraphCast AI sets new standard for weather forecasting accuracy
• Samsung breaks language barriers with real time AI call translations
