AI isn’t human, it’s math
AI is very broad, meaning there’s no singular definition, meaning contradictory claims reasonably coexist
AI is very broad, meaning there’s no singular definition, meaning contradictory claims reasonably coexist. Below are some corrections.
AI is difficult to detect the use of. Few people know when they’re interacting with AI, last surveyed only 27%. But AI is everywhere, from email spam filters to your iPhone keyboard. Generative AI brings AI central to the interface, primarily used by super users under 40. Other AI systems are quietly integrated into life: recommenders, computer vision, self-driving cars, autonomous robots, etc.
AI has become extremely capable, therefore many claims about AI exist, eliciting hype and fear: AI is making entry level jobs obsolete, AI agents are taking over computer tasks done by humans, AI will be smarter than humans, 400 million workers could be displaced, but AI can create 97 million new jobs.
POVs are influenced by (1) AI products & experiences, which anyone can have a perspective on, and (2) knowledge about how AI systems work, which few people besides Data Scientists have deep expertise in
Combined, fundamental truths become more clear:
- AI simulates human intelligence
An aspiration, not how it works - AI is technology
A concept, not a product - AI is a tool
AI is a hammer searching for nails, AI as a design material - AI is a surrogate of data
AI is really a data innovation - AI is a partner
Co-pilots, agents, collaborators - AI is a new interaction paradigm
The first in decades
1. AI simulates human intelligence
AI is often compared with humans. This analogy makes a complex system feel tangible. It has also influenced AI products, per Daniel Warfield: The only reason AI can do things humans can do is because they’re designed to mimic people.
But AI isn’t human; it’s fundamentally different. AI is data, math, statistics & complex systems. It’s designed to complete tasks, and accomplishes them with variable accuracy.
A crucial distinction: AI’s quest to replicate human-like intelligence in machines is an aspiration, not how it works
Humans are incredibly complex. AI is programmed to complete certain tasks humans happen to already be doing.
AI is good at some tasks. Humans are good at other tasks. Human and AI will be good at a new category of tasks we are figuring out, but it takes knowing what AI is uniquely good at, along with appropriateness, including short and long term consequences.
Therefore, human vs. AI raises questions about what it means to be human, whether we should be partnering with AI, competing against AI, or accelerating AI displacement, in search of better jobs for people.
Humanists believe in designing AI partnerships, more in point 5.
2. AI is technology
AI has been around over 60 years, a term coined in 1956 at the Dartmouth Conference. It was first named “Artificial Intelligence” as a curious, trendy term to generate hype and obtain funding. Similar to how it’s used today.
History shows every new technology tends to replace a human task. Same with AI. Because it’s easier to think of things people already do, than things people don’t or can’t do, but machines can.
Technology and AI are categories. Saying something is an AI product is like saying something is a technology product.
People don’t remember definitions. They remember stories, abstractions, metaphors, products.
What’s the story of AI? Now is the opportunity to define.
3. AI is a tool
Tools are instruments to carry out a certain function. AI is a metaphorical tool. Like duct tape, AI is good for many things, not everything. AI can also be designed to be a tool, helping people accomplish certain tasks.
We shape our tools, and thereafter our tools shape us, John Culkin
a) AI is a design material
In most cases, AI is already chosen as the underlying mechanism, if the intention is to build an AI product. Ideation becomes more limited because we’re starting in the middle of the design process where technology medium has already been selected.
Designers then come up with product experiences with AI, and like paint on canvas, AI is used as a material for design. Designers study its qualities & make usability choices, reflecting in action.
Consider the growing number of AI tools as a result, which really are products that embody AI to accomplish tasks for which it was designed.
b) AI is a hammer searching for nails
They say AI is a hammer searching for nails. AI is ready for action but needs to be paired with the right use case for what your AI can reasonably do.
The match between technology and the right user need to solve for may come months or years later when the AI capability we’ve dreamed about becomes possible & reliable.
We shape our AI tools; our AI tools shape us.
4. AI is a surrogate of data
AI systems are as good as the data they’re trained on. AI systems got good because the quality and quantity of data got better. From a colleague:
Today’s AI is really a data & data modeling innovation
Over the past few years, AI has become significantly more capable: “partly due to better programming and algorithm development… 90% thanks to the fact that AIs have been trained on significantly larger datasets.”
AI systems perform best when they reach a certain threshold. Less data may not be representative, and the model can’t recognize edge cases. More data costs significant computational power, sacrifices speed, only to marginally improve the model’s accuracy. From Will Lockett’s technical perspective, it reaches a point of diminishing returns.
From a responsible AI perspective, data reflects biases of (1) people who collect the data, whether data collected is inclusive & diverse representing industries, use cases or cultures, (2) data availability, what datasets are licensed for use in a secure, permissible, privacy-preserving way, and (3) data cleanliness e.g. audio data is most cumbersome to obtain in the wild and difficult for training good models, often containing background noise, volume discrepancies, incomplete corrupted sound bytes, etc.
5. AI is a partner
AI is a partner, collaborator, co-pilot, whichever term you like.
Partnerships are well exemplified in sports teams:
In any good partnership, each party watches out for each other party, and is able and willing to intervene when one of them needs help. (Laura Major, Julie Shah)
AI systems need to know what the person is trying to do, in order to know when to helpfully step in. People also need to know what AI systems are trying to do, to actively anticipate when to get involved.
This overarching human-AI partnership concept covers both automation and augmentation, even empowerment. Sometimes, AI systems are best positioned to accomplish the task on their own. Other times, AI augments humans to help accomplish things better or not possible before.
Some situations may require supervision, but threshold of trust & transparency is up to human discretion. If people trust the system less, supervise and be ready to take over the task. Design to index on human decision making, with AI as an option. If people trust the system more, manage and let it be. Give people system status updates as needed.
Both work streams can be possible, supporting the variable threshold of trust that goes up and down, dependent on situation & task.
6. AI is a new interaction paradigm
Engineers create new technology that allow new capabilities. Designers do not invent new technologies. Instead, they craft novel assemblies of known technologies into products (Louridas, 1999), including the interfaces that humanize them.
With the new AI systems, the user no longer tells the computer what to do. Rather, the user tells the computer what outcome they want, Jakob Nielsen
Interaction is communication & feedback. And feedback loops are the cornerstone of any good partnership. Telling the computer is direct feedback. Designing the computer to adapt and learn from new inputs and data is another. All other non-verbal, non-language based interactions are a third, some examples:
Liking a post means you may want to more related content. Skipping a song means you want less of this type of music. Traveling the same route every day despite traffic means you prefer familiarity over time to destination.
AI may want a communication-first design paradigm, but communication is nuanced. There are many types of interactions AI can enable, not all of them have to be visible to be valuable.
In theory, AI is defined as simulating intelligent human behavior, imitating decision making, completing tasks that typically require humans.
In practice, AI is a category, an aspiration open to many possible realities, achieved with data & math.
