Definition is important when talking about AI

Artificial intelligence is everywhere these days. It's around news, podcasts and any water cooler. The new and talked about term artificial general information (AGI) dominates the conversation and raises more questions than answers.

So, what is AGI? Will it replace a job or unlock a solution to the world's biggest challenges? Does it fit into social priorities or redefine them? Is our future more “Jetson” or “Terminator”? The answer is not easy. They rely on how AI is defined and what it does.

As the first director of the AI ​​Learning Center, Joe Sutherland, Emory's community hub for AI literacy and a core component of the AI.humanity initiative, is unfamiliar with difficult questions about AI. Last summer he embarked on a statewide tour to measure AI and develop Georgians with skills to thrive in a technology-focused future. A diverse audience consisting of experts, business owners, lawmakers and students shared many of the same core questions.

In this Q&A, Sutherland answered the most pressing concerns about AGI and helped her get through the hype and find hope.

What is the difference between artificial intelligence, artificial general information, artificial super intelligence?

The definition changes over time, which causes some confusion. Artificial intelligence is not a monolithic technology. This is a set of technologies that automate tasks and mimic decisions that humans normally do. They delegate the authority to make these decisions to the machine.

Traditionally, when people talked about artificial general information (AGI), they meant Skynet for the “Terminator” or HAL for the “2001: Space Odyssey,” a machine with free will and approximate human abilities.

Today, several major labs mean computer programs that redefine AGI and can perform better than experts on specific tasks.

Artificial superintelligence (ASI) is a modern term for what we called superintelligence or singularity. That's what we thought of at the same time as AGI, a humanoid robot that surpasses human intelligence.

Do you already have AGIs?

It depends on the definition used. Yes if you are using task-oriented definitions from the lab. AI is good at getting information and summarizing it in a way that human-rated people say, “Oh, this is pretty good.”

Large-scale language models (LLMs) like ChatGpt are superior to those trying to enter medical school on the MCAT. But that's not true intelligence. It's like giving Google to a student during an exam. True AGIs need to show inference as well as information search and pattern matching.

What is the difference between reasoning and what AI is doing today?

Today's models give the impression they are reasoning, but they simply investigate the information successively and summarise it. They don't understand the world. Just predict the next word based on the pattern. When tested with actual inference tasks like the Tower of Hanoi or logic puzzles, LLM often fails unless you remember the answer.

Humor is another example of lack of AI. From a humanities perspective, humor lies at the intersection of comfort and discomfort. That boundary always shifts. Chatbots only reflux what they have seen in the past. They don't understand the boundaries. True humor is something they can't do.

Similarly, companies often do not have data showing broader trends that are happening “outside” of the company. AI models trained with internal data cannot integrate where we went. This requires reasoning, intuition and value adjustments.

So how close are we really?

If you use the old definition of AGI, like “Terminator,” I think we're far away. LLM does not have inference or intuitive creative abilities, so it does not bring us anywhere. It does not provide a framework for efficiently discovering new information. If you want to get closer, you need to develop an entirely new architecture.

One step in the right direction is to embed a predictive architecture, or JEPA. Instead of stringing words like LLM, we derive deeper relationships between concepts and activate those inferences to achieve higher level goals.

It's refreshing to know that throwing all the encyclopedia entries of society to LLM does not produce human-level intelligence. Of course, I'm bringing it to a boil. There is more to humanity than to make eye contact.

What are AGI's biggest promises and dangers?

The technology today is amazing. They allow them to perform tasks in the hours they had to spend days previously. The promise is efficiency. It is a tool that can help you summarize your research, assist in medical diagnosis, and plan your shopping list for the week. It helps people make more money, spend more time interacting with their families and hobbies, and lead a longer, healthier life.

Danger is not technology. It is a lack of public understanding. Because AI literacy is needed, people understand that AI is being used in the right way and not.

What oversights and guardrails do you need?

The key is to provide a framework that balances these models: dataset-fueled intellectual property with the value of what is built.

Some companies have argued in court that cutting people's data without consent or compensation is justified in order to advance society. It's a manipulative and troublesome argument.

If the needs of those who contribute to these technologies are represented and they are properly rewarded, it will encourage greater innovation and use. In addition, thorough testing is required to determine whether these models break biased results or where they occur.

Can AI be tailored to human values?

It's more of a social question than a technical question. Last summer, I spoke about Emory's statewide AI workforce development tour and the Rowen Foundation and Georgia Chamber of Commerce. One audience member asked, “How can we ensure that the AI ​​models we are building have American value?”

What makes America special is that we value the opportunity to challenge. We never agree with everything, so the bigger question is: How do we create a system with responsive guardrails that adapt to evolving social values? What framework allows us to still have a robust argument and ultimately lead to a good decision?

These questions are not new to the advent of AI. We have been asking them for centuries. However, the rapid rise of technology and its increasingly centralized power forces us to revisit how we approach collective action.

Once AGI is right, what will the best version of the future look like?

I think the best version of the future with all these technologies is something that will free us to do more meaningful work and reduce the time we spend doing things that we don't enjoy. This means more time for innovation and creative problem solving.

Emory's AI research has already transformed healthcare, leading to improved diagnosis and treatment of diseases such as cancer, heart disease and diabetes. Textual analysis is used to uncover patterns of public policy that make governance more efficient and fair. Our scholars are also looking at ways in which AI protects people's rights and grows their businesses.

AI offers more advantages than the drawbacks if they empower people through education and include them in conversations so that they can advocate for themselves and those they care about.

If deployed thoughtfully, these technologies can amplify human potential rather than replace it.