Inching towards AGI: How reasoning and deep research are expanding AI from statistical prediction to structured problem-solving

Artificial intelligence has evolved at an amazing pace. What seemed to be a science fiction a few years ago is now an undeniable fact. In 2017, my company launched an Amnesty International Excellence Center. Artificial intelligence certainly was improving in predictive analyzes, and many ML learning algorithms (ML) was used to identify sound, discover random mail, spelling examination (and other applications) – but it was early. At that time, we believed that we were only in the first half of the artificial intelligence game.

The arrival of the GPT-3, especially the GPT 3.5-which was seized to use the conversation and was the basis for the first Chatgpt in November 2022-exciting turning point, now remember it forever as “ChatGPT”.

Since then, there has been an explosion of artificial intelligence capabilities from hundreds of companies. In March 2023, Openai GPT-4, which promised Sparks of Aoy (artificial intelligence). By that time, it was clear that we were outside the first half. Now, it looks like we are in the final extension of a completely different sport.

AGI flame

Two years later, AGI flame began to appear.

In a recent episode of The Hard Fork Podcast, Dario Amodei – who was in the industrial intelligence industry a decade ago, previously said as a deputy deputy in Openai, and now the CEO of Antarubor – there is an opportunity from 70 to 80 % to have a “very large number of artificial intelligence systems” that have become almost more than humans in everything before the end of December, and guess 2026 or 2027 “

Evidence for this prediction has become more clear. Late last summer, Openai O1 – the first “thinking model” launched. They have since released O3, and other companies have put forward their thinking models, including Google and Deepseek. Exiles use the idea chain (COT), which breaks down complex tasks at the time of operation to multiple logical steps, just as a person may approach a complex task. Sophisticated artificial intelligence agents, including deep research from Openai and Google Ai, recently appeared, as they have made tremendous changes on how to make research.

Unlike the previous large language models (LLMS) that mainly coincide with training data, thinking models are a fundamental shift from statistical prediction and organized problem solving. This allows Amnesty International to address new problems that exceed its training, allowing real thinking rather than identifying advanced patterns.

I recently used a profound search for a project and was reminded of the quotation from Arthur C. Clark: “Any technique is advanced enough that cannot be distinguished from magic.” In five minutes, this artificial intelligence produced what would take 3 to 4 days. Was it perfect? No. Was it close? Very yes. Soon these factors are really magic and transforming and are among the first strong factors that will soon come to the market.

The most common definition of AGI is a system that is able to do any cognitive task that a person can do. These first agents of change indicate that Amodei and others who believe that we are close to this level of artificial intelligence may be true, and that AGI will be here soon. This reality will lead to a great deal of change, and requires people and operations to adapt in a short time.

But is it really Agi?

There are many scenarios that can appear from the arrival of proximity to strong artificial intelligence. It is difficult and frightening that we really don’t know how this will happen. New York Times The column writer Ezra Klein addressed this in a modern podcast: “We are raging towards AGI without really understanding what this means or what this means.” For example, he claims that there is a little critical thinking or emergency planning that occurs around the effects, for example, what this really means for employment.

Of course, there is another perspective on this uncertain future and lack of planning, as shown by Gary Marcus, who believes that deep learning in general (and LLMS specifically) will not lead to AGI. Marcus released up to a decrease in Klein’s position, pointing to noticeable deficiencies in the current artificial intelligence technology, indicating that we are likely to be far from AGI.

Marcus may be correct, but this may also be just an academic conflict over the indications. As an alternative to the term AGI, Amodei simply refers to “strong AI” in its devices from the Loveing ​​Grace Blog, because it transmits a similar idea without an inaccurate definition, “science fiction and noise luggage”. We call it what you want, but artificial intelligence will only grow more powerful.

Playing with fire: The potential futures for Amnesty International

in 60 minutes An interview, the CEO of Alphabet Sundar Pichai said he believes that “the most depth of human technology is working on it. More deep than fire, electricity or anything we did in the past.” This is definitely suitable for the increasing density of artificial intelligence discussions. The fire, like artificial intelligence, was a discovery that changed globally, which strengthened progress but demanded the control of the prevention of the disaster. The same exact balance applies to artificial intelligence today.

Discovering the tremendous power, the fire turns civilization by enabling warmth, cooking, minerals and industry. But it also brought the destruction when it is not controlled. Whether artificial intelligence becomes our greatest ally, or if the retreat of us will depend on the success of its fire. For more metaphor, there are different scenarios that can appear soon from the most powerful artificial intelligence:

1. The controlled flame (Utopia): In this scenario, artificial intelligence is harnessed as a power of human prosperity. Productive Skyrockkets, new materials are discovered, personal medicine becomes available to everyone, goods and services become abundant and inexpensive and individuals are free from toil to follow work and most important activities. This is the scenario that many acceleration specialists have defended, as artificial intelligence achieves progress without being seduced in a lot of chaos.
2. Unstable fire (challenge): Here, artificial intelligence brings indisputable benefits-a revolution in research, automation, new capabilities, products and problem solving. However, these benefits are unevenly distributed – while some flourish, others face displacement, expand economic divisions and emphasize social systems. Wrong information and security risks are spread. In this scenario, society is struggling to balance the promise and risks. It can be said that this description is close to the current reality.
3. Washing fires (dysmenorrhea): The third path is a disaster, and the possibility is associated with the so -called “Domers” and “the possibility of Doom”. Whether through unintended consequences, reckless publishing or artificial intelligence systems that go beyond human control, artificial intelligence actions have become not verified, and accidents occur. Trust in fact erodes. In the worst scenario, artificial intelligence is out of control, threatening lives, industries and entire institutions.

Although each of these scenarios seems reasonable, it is exciting to upgrade that we do not really know which one most likely, especially since the schedule may be short. We can see early signs for each: increasing the productivity that AI drives, and the wrong information that is widespread, and erosion of confidence and concerns about the annoying models that resist handrails. Every scenario would cause it to adapt to individuals, companies, governments and society.

Our lack of clarity on the course of the impact of artificial intelligence indicates that some of all three future decades is inevitable. The ascension of artificial intelligence will lead to a paradox, nourish prosperity with unintended consequences. Amazing breakthroughs will occur, as will accidents. Some new fields will appear with confusing possibilities and work prospects, while other strong economies will fade to bankruptcy.

We may not have all the answers, but the future of strong artificial intelligence and its impact on humanity is now written. What we saw was at the top of the Paris Ai movement is a mentality in the hope of getting the best, and it is not a smart strategy. Governments, companies and individuals must form the path of artificial intelligence before we form. The future of artificial intelligence will not be determined by technology alone, but through the collective options we make on how to publish it.

Gary Grossman is EVP for technology in Edelman.