Tasteful AI

I want to change the dialogue around Artificial Intelligence. Here are some thoughts about how I would know I was successful at changing this dialogue:

We would be talking about AI built in a way that “taps the power of limits.”

• We can develop AI that learns from limited data and reduce or eliminate the need for massive datasets.
• We can incorporate constraints that reflect ethical, social, and environmental considerations.
• We will be talking about resource-efficient algorithms, optimizing the use of computational resources, and reducing energy consumption and processing time.
• We would consider adaptability — designing AI systems that can adapt to different contexts and environments because they harness the power of understanding that context.
• Finally, we would be talking about building AI that encourages collaboration and is inspired by natural systems and processes to mimic the efficiency and resilience of nature.

We could call this “tasteful” AI.

• AI, which has manners, uses please and thank you.
• AI that understands different cultures and is taught to be curious and polite.
• AI that shows its source data.
• AI that shows evidence for its prediction.
• AI, whose human creators value where its data comes from, provides, in the form of attribution, who has written the words or the program or created the art used to generate the variations or transformations.
• AI, whose human creators are held responsible for the environmental impact of the compute consumed during the AI creation.
• AI that is created with the intent to augment a human.

I am confident that showing your references, showing your work, and demanding that AI be able to explain what data created what prediction and why — this will become a norm, and this must become a norm. This type of attribution would follow the natural understanding that if something (data) is given, something equivalent is expected. For example, consider this quote from the Māori Data Sovereignty and Digital Colonisation.

“It is a customary belief that when you share knowledge with people, that the person who you are sharing information with, then acquires a spiritual part of you. From a western perspective, if you imagine a thought of a person in your mind, you have no control of that thought existing. No one else can see it, but you know it is in your brain and a part of you. Therefore, in a customary Maori perspective that Māori Data contains wairua, mauri, it becomes a form of genealogy of whakapapa and therefore becomes sacred or tapu.”

Data is an artifact of a human experience, it is something given. It is something that we humans have been sharing in our digital exhaust since time began, as our breath laughter and as our cry that contains the duality of: We belong and We are special. Data is what is being harvested without our consent, no wonder our world feels so out of balance. There is a rule of reciprocity that must be followed. “Tasteful” AI gives attribution, it is something that can be made to help humans remember where an idea comes from, where words and thoughts are remixed and made relevant to the next generation.

AI can explain and attribute the humans who created its sources. This is explainable AI, and this type of explainable AI will be easier to debug and understand and far more valuable than the brute force models we see today, which are being replicated for “competitive advantage.” I wonder when the scientists will say “uncle” STOP.

Stop creating larger models with more and more parameters — doing the same thing and expecting a different result — the definition of insanity. The pile of statistics without grounding (context — documenting and attributing) will never render an intelligence that can understand the meaning of the language — in the context it was given.

“Understanding is not an act. It is a labor.”

To point out the obvious — there is a phenomenal economic incentive for “tasteful AI.” The ability to provide provenance and lineage increases the value proportional to the quality of the origin story. Look at how the big auction houses have created the ability to sell things for a great deal more by providing the origin story. Perception is the rule. There is a need to have AI be portable, small enough only to use the energy on a smartphone and to be carried into the far corners of the world, augmenting humans to explore the depths of the seas or the vast universe that is still so unknown.

So Tasteful AI taps the power of limits, is built on the efficiencies learned from our natural world, can explain itself, and tell the stories of where it got the data it uses to make the prediction. Tasteful AI can be done with “small data” and sound engineering. It can be made sustainable and portable. We could see that Tasteful AI would combine all the “tools,” using the technology to augment humans and unite and build communities.

I like this worldview.

I want tasteful AI, please.