OUR GUIDING IDEA
Better together. Humans in charge.
AI brings capability. People bring purpose, judgment and accountability.
The combination earns “superior” when it produces a better result.
THE CASE FOR USEFUL AI
Positive stories. Measurable progress.
We should recognise a useful result when the evidence supports it. These two studies show different kinds of progress: helping people at work and improving a scientific prediction.
AT WORK · RESEARCH PUBLISHED IN 2025
AI assistance + support agents
A study of 5,172 customer-support agents found that access to an AI assistant increased issues resolved per hour by 15% on average. Less experienced and lower-skilled workers improved both speed and quality. The most experienced and highest-skilled workers saw small speed gains and small quality declines.
The partnership: an AI assistant supported people doing the work. The 15% average belongs to this studied setting; results varied across workers.
Source: Brynjolfsson, Li and Raymond, Generative AI at Work, The Quarterly Journal of Economics (2025). Open author manuscript.
IN SCIENCE · RESEARCH PUBLISHED IN 2021
AI predictions + scientific investigation
In the CASP14 assessment, AlphaFold produced protein-structure predictions much more accurately than competing methods. The researchers reported accuracy approaching experimental structures in a majority of cases.
The assessment tested predictions against structures that had been experimentally determined but were not yet publicly disclosed. Human experimental work supplied the reference for judging the predictions.
Our takeaway: AI contributes a prediction; scientists investigate what it means and whether it is sufficient for the intended use. That is how a stronger tool can expand human capability.
Source: Jumper and colleagues, Highly accurate protein structure prediction with AlphaFold, Nature (2021).
Useful capability deserves recognition. The four cases that follow show why judgment and the ability to correct mistakes still matter.
THE HUMAN PART OF THE EQUATION
Judgment makes the difference.
AI + human judgment = superior intelligence is the idea behind this page. We want tools that help people achieve more, with people choosing the goals and remaining accountable for how those tools are used.
The equation is a standard to work towards. The positive studies show specific gains; the cases below show what can go wrong when a confident answer is accepted or an automated service cannot put things right.
Sometimes the human contribution is the question nobody asked, the evidence somebody checked, or the decision to stop. That is why “What if the superior intelligence is us?” still matters.
The chatbot gave an answer.
A person challenged the outcome.
The human contribution: persistence and evidence
After a family death, Jake Moffatt relied on Air Canada’s chatbot, which said a bereavement fare could be requested after travel. The airline’s actual policy did not allow that. Moffatt pursued the discrepancy, using a screenshot of the chatbot’s answer.
British Columbia’s Civil Resolution Tribunal found Air Canada responsible for negligent misrepresentation and awarded damages. The company could not avoid responsibility for information on its website because a chatbot supplied it.
What the human did
Moffatt kept evidence and challenged the refusal. The tribunal weighed the record and provided a remedy. Here, human judgment supplied accountability that the automated exchange had not delivered.
The decision does not identify the chatbot’s underlying technology. This is a documented automation failure, not proof that a particular large language model failed.
Source: Moffatt v. Air Canada, 2024 BCCRT 149, especially paragraphs 14–32.
The cases sounded convincing.
They did not exist.
The human contribution: checking the original
In Mata v. Avianca, lawyers submitted court filings containing fabricated judicial opinions generated by ChatGPT. The opposing lawyers could not locate cited authorities. The court investigated and confirmed the problem.
The court imposed sanctions after the lawyers continued to stand behind the false material. It also said using a reliable AI tool for assistance was not inherently improper: the lawyers still had to ensure their filings were accurate.
What the human did
Other lawyers and the court tested the citations against the actual legal record. They treated a polished answer as a claim to verify.
This case also contains human failure: people submitted and defended the fabricated material. The advantage came from verification, not simply from being human.
Source: The court’s sanctions opinion (PDF), pages 1–6.
A superhuman player.
A very human opening.
The human contribution: learning an unexpected strategy
Researchers found a weakness in powerful Go programs: certain circular groups of stones could make them badly misjudge the board. Amateur player and researcher Kellin Pelrine learned the strategy and used it to beat superhuman versions of KataGo and Leela Zero without assistance during play.
Another AI helped discover the strategy. This was a human learning from an AI-assisted investigation, rather than independently outplaying those programs through ordinary Go skill.
What the human did
Pelrine understood and applied a strategy that exploited the opponent’s blind spot. A program could be stronger in typical games yet lose in a situation that exposed its weakness.
The result demonstrates a specific vulnerability in the tested systems. It does not mean humans generally became stronger Go players than AI.
Sources: The ICML 2023 paper and the researchers’ account of the human games.
The answer missed the joke.
People did not.
The human contribution: recognising context
Following the launch of AI Overviews, Google acknowledged errors involving satire and sarcastic forum posts. Its own explanation described a rock-eating query and an overview suggesting glue to keep cheese on pizza.
Google said it made changes to better detect nonsensical queries and limit misleading use of forum content. It also noted that many circulating screenshots were fake; the examples here come from Google’s acknowledgment.
What the human did
People flagged answers whose fluent presentation hid an obvious problem. Our reading: recognising a joke, questioning a source and noticing an absurd premise can matter more than producing an immediate answer.
These were documented failures from the 2024 rollout. They are not a claim that today’s system still gives those responses.
BEYOND A SCOREBOARD
Could the superior intelligence be the partnership?
Finding an AI mistake does not prove that every human would have avoided it. People also overlook evidence, repeat false claims and make poor decisions. The useful question is what improves the result.
The Go example makes this especially clear: an AI-assisted discovery became a strategy a person could understand and use. The court example shows the reverse possibility: a tool’s mistake grew more harmful when people accepted it without checking.
For us, a better partnership gives people the means to inspect an answer, correct it and decide when the tool should not act. Human review needs time, knowledge and the authority to change the outcome. A person clicking “approve” is not enough.
OUR POINT OF VIEW
AI can help us do better. People decide what better means. We believe people should choose the goals and limits, have a meaningful say in decisions affecting them, and hold the organisations using these tools accountable.
Being better at a task does not settle who should have authority over it. A performance comparison does not measure a person’s worth. And superintelligence remains a separate research concept about much broader cognitive capability.
THE CANADIAN TEST
Human control has to be real.
For a Canadian resident using a public service, a worker checking a draft or a customer disputing an answer, “a human is in charge” should mean something practical. This is the standard we would ask organisations to meet:
- People set the purpose.Define what the system is allowed to do, what success means and which decisions need a person’s approval.
- People can check the answer.Make evidence, uncertainty and important limits understandable to the people relying on it.
- Someone can pause, correct or override it.Give the reviewer enough time, knowledge and authority to change the outcome.
- Affected people can challenge a decision.Provide a clear route to a person who can investigate and help. Make it usable across the languages and accessibility needs of the people served.
- An organisation owns the consequences.Name who is responsible for fixing mistakes and reviewing whether the system is delivering its promised benefit.
Superior intelligence should make us more capable—and keep us in charge. That is the purpose we want our name to serve.
THE NEXT HUMAN CASE
Where did a person make the difference?
A tool that helped you do better. A mistake you caught. A decision you challenged. Tell us what happened and share a source if there is one.
The next example
could be yours.
Pass the question on.
Explore AI, SI and the naming debate · Choose what better means to you