What is Human AI?

For a technical treatment of Causal Economic Machine Learning, visit the article in the journal AI at www.mdpi.com/2673-2688/5/4/94.

Human AI refers to the convergence of artificial intelligence and human intelligence, where AI augments humans and mirrors the science of HI. These are the two pillars that define Human AI. It comes down to leveraging the power of machines to process massive amounts of data, find relationships and combining that with models of realistic human decision making in the face of real world constraints. Unfortunately, academic research in AI and HI have for the most part been evolving in almost complete silos from each other. If we believe human intelligence is effective in our world, we must bring these together into Human AI. In a practical sense this comes down to the way data is structured and then optimized in algorithms.

There are a few major approaches that explicitly ground Human AI in human intelligence.

Human-Centered AI (HCAI) and Humanistic AI

Human-centered AI (HCAI) is a branch of AI aimed at improve human capabilities, using AI to augment, rather than replace human abilities. HCAI models generally attempt to deliver the highest possible level of benefits to humans, including elements of social fairness. Humanistic AI is a related term coined by Tom Gruber, founder of Siri, to illustrate the need for AI to augment human intelligence, directly and through collaboration. The concept of augmentation central to these approaches make them important elements of Human AI, aligning to its first pillar.

AI Clones

Innovative companies already exist that will help create customized digital clones of a human’s voice, visual appearance, knowledge and personality. AI cloning of humans generally fits within the category of Human AI, since it literally mirrors a human and is created to augment them. The thinking, reasoning and actions built into AI clones is directly aligned to the approach of Human AI. But to the extent that a clone becomes a completely independent entity, thinking and acting independently of the direction and supervision of its human owner, it breaks the definition of Human AI.

Causal Economic Machine Learning

The term Human AI was introduced in an academic paper I had the pleasure of writing, published in the journal AI in 2024, which introduced Causal Economic Machine Learning (CEML). But the concept of Human AI is broader than CEML. It includes the areas of Human-centric AI, Humanistic AI and Human AI Cloning mentioned above. CEML introduces a structural model that combines the power of causal machine learning AI with the realistic human behavioral model (HI) of causal economics. CEML and AI Clones are the two areas of Human AI that most explicitly build HI right into the Human AI model. For a technical treatment of CEML, visit the article in the journal AI at www.mdpi.com/2673-2688/5/4/94.

Four Levels of Human AI

Similar to the way in which AI has three core levels—narrow AI, AGI and SGI—Human AI has four core levels. These levels, in increasing levels of sophistication are:

• Information AI
• Task AI
• Decision AI
• Agent AI

Information AI

This is the big area of action today, represented by the massive presence of generative AI. It’s about generating and editing content and it’s helping in many areas of life.

Task AI

This level is exactly as it sounds. AI is trained to perform assigned tasks for its human owner. This includes very popular AI assistant applications in use today. It is distinguished by a complete lack of decision autonomy. The AI is directed what to do and it does it.

Decision AI

This stage involves real-time use of AI to make personalized predictions of outcomes based on a decision maker’s choice. It takes into account the knowledge and personality of the decision maker and can predict realistic outcomes, and variance from goals, with corrective action plans. It uses AI cloning technology and CEML modelling to provide 100% personalized predictions for people, in place of the generic guidance that exists in the generative AI tools of today.

Agent AI

AI agents sit at the top of the Human AI hierarchy. These are essentially ‘AI representatives’ that go out and make some decisions and take some actions based on a person’s pre-set and ongoing criteria and supervision. Think of AI employees, AI real estate agents etc. They can have their own ‘personalities’ and can also be set up to clone significant elements of their owner’s knowledge and personality in order to represent their human owner effectively in dealings. This area heavily involves AI clone technologies. It is all about dispatching ‘virtual versions’ of humans that can do assigned things, where these versions always ‘report up to’ a human that is responsible for them.