Optimism in the face of uncertainty

In June 2026, my MSc advisor, Joëlle Pineau, received an honorary doctorate from the University of Ottawa and delivered the commencement address to the graduating class of its Faculty of Engineering. Joëlle is a professor at McGill, previously led Meta’s Fundamental AI Research organization, and is now Chief AI Officer at Cohere.

Her address had a title that immediately caught my attention: “Optimism in the Face of Uncertainty.”

It sounds like a piece of general life advice. But the idea came from somewhere much more precise: mathematics.

Joëlle described a principle she encountered while studying how intelligent agents should make decisions when they do not know exactly what will happen next.

When faced with several possible actions, the obvious strategy is to choose the one that currently looks best.

But that is not always the best strategy.

Sometimes the most interesting action is the one whose outcome is uncertain, but whose upside could be much larger.

This principle is known as optimism in the face of uncertainty.

And the more I thought about it, the less it seemed like a lesson about reinforcement learning and the more it seemed like a useful way to think about research, careers, and life.

A mathematical reason to be optimistic

Imagine that an agent has several actions available to it.

One option has been tried many times. Its reward is reasonably good and reasonably predictable.

Another has barely been explored. Its average estimated reward might look lower, but because the agent knows so little about it, there is considerable uncertainty around that estimate.

A purely greedy agent chooses the first option.

An optimistic agent asks a different question:

How good could this unexplored option plausibly be?

This basic idea appears throughout reinforcement learning and the study of multi-armed bandits. Algorithms such as Upper Confidence Bound (UCB) deliberately account for uncertainty when deciding what to explore. An action can become attractive not merely because its estimated reward is high, but because there remains a plausible chance that it is substantially better than we currently believe.

There is an elegant logic behind this.

If the optimistic estimate turns out to be right, you discover something valuable.

If it turns out to be wrong, you still gain information.

As Joëlle put it in her commencement address, when reality falls short of your optimism, you have not necessarily lost: you have gained knowledge.

That is a subtle but important distinction.

Optimism here does not mean convincing yourself that everything will work out.

It means recognizing that uncertainty can have value.

The same problem appears in our careers

The connection to life becomes especially clear around graduation.

Suddenly, there is an enormous decision tree in front of you.

Which job should you take? Which city should you move to? Should you stay in academia? Join industry? Start something? Work on the established problem where you know you can succeed, or the strange new question where you have no idea what will happen?

We often respond to these decisions by trying to eliminate uncertainty.

We gather more information. Make spreadsheets. Compare salaries. Ask more people for advice. Try to predict exactly where each path will lead.

Some of this is obviously sensible.

But eventually there is a limit to how much more thinking can tell you.

At that point, uncertainty is not something that can be reasoned away. It has to be experienced.

You have to choose an action and observe what happens.

That is why one of my favourite lines from Joëlle’s address was her description of a degree as a “license to take risks.” She encouraged the graduates to say yes to the project that seems too ambitious, the role they do not quite feel ready for, or the question nobody in their field has thought to ask.

I think that is particularly good advice early in a career.

When you are young, the benefits of exploration have a long time to compound.

A project that goes nowhere might introduce you to an entirely new field.

A job that turns out not to fit might clarify what kind of work you actually want to do.

An unsuccessful application might lead you to meet someone you otherwise never would have spoken to.

A research question that fails might reveal a much better question hiding underneath it.

The outcome itself is only part of the return.

The other part is what the experiment teaches you.

Optimize for information gain

We tend to evaluate past decisions with an overly simple scoring system.

The paper was accepted: good decision.

The paper was rejected: bad decision.

The startup succeeded: good decision.

The startup failed: bad decision.

I got the position: good decision.

I didn’t get it: bad decision.

But that confuses the quality of a decision with the outcome of a single trial.

Suppose you spend three months exploring an ambitious research idea and eventually conclude that it does not work.

You could call those three months wasted.

But perhaps during those three months you learned a new method, became familiar with a literature you had never encountered, met several collaborators, discovered why an apparently promising idea fails, and stumbled across a different problem that turns out to be much more interesting.

The experiment failed.

The information gain did not.

That suggests a question worth adding whenever we face an uncertain opportunity:

What will I know after doing this that I don’t know now?

This changes how many opportunities look.

Something with only a moderate probability of immediate success may still be an excellent bet if:

• the downside is limited,
• the potential upside is large, and
• failure will teach you something useful.

That is a very different criterion from simply choosing the path most likely to succeed.

Run cheap experiments

There is, however, an important caveat.

The most attractive experiments are often the opposite: small downside, large upside, fast feedback.

• Send an email to someone whose work you admire.
• Apply for the position you think you are slightly underqualified for.
• Spend a weekend building the prototype.
• Try the new research direction for two weeks.
• Submit the paper.
• Take the difficult course.
• Write the essay.
• Go to the conference.
• Ask the question.

These actions can fail.

But most failures are cheap.

And occasionally one of them changes the trajectory of your life.

Perhaps the better rule, then, is not simply take more risks.

There is an asymmetry to opportunity

Another reason this framing is useful is that many interesting opportunities have asymmetric payoffs.

Imagine emailing a researcher you admire.

Maybe there is a 90% chance nothing happens.

That sounds like a terrible bet if you evaluate it by probability of success alone.

But suppose the cost of sending the email is ten minutes, while the 10% outcome could produce a collaboration, mentorship, job opportunity, or entirely new research direction.

The expected value begins to look very different.

The same principle applies to applications, introductions, side projects, conferences, writing publicly, and ambitious research questions.

Humans often overweight the probability of embarrassment or rejection and underweight the magnitude of the upside.

We see the likely outcome:

They probably won’t respond.

We pay much less attention to:

But what if they do?

Optimism in the face of uncertainty forces us to consider both.

Research itself depends on this mindset

The idea feels particularly natural in science.

Good research begins with uncertainty.

If we already knew the answer, there would be little reason to run the experiment.

Yet researchers can gradually become surprisingly good at avoiding uncertainty.

We learn what reviewers like.

We learn which analyses are predictable.

We learn which questions can produce another incremental publication.

And those incentives can push us toward exploitation rather than exploration: repeatedly mining areas where we already know rewards are available.

But some of the most interesting research exists precisely where the confidence intervals are widest.

A strange result.

A new dataset.

A methodology borrowed from another discipline.

A question that sounds slightly ridiculous when you first say it aloud.

Most of these directions will not produce breakthroughs.

That is the point.

Exploration is valuable because you cannot know beforehand which one will.

Open science as an optimistic bet

Joëlle also connected this philosophy to another theme that has defined much of her career: open science.

For years, she has pushed machine learning toward greater reproducibility and more transparent research practices. Her work helped establish reproducibility initiatives at NeurIPS, including code submission practices, reproducibility challenges, and checklists for reporting machine-learning experiments. In a 2021 Journal of Machine Learning Research paper, she and her co-authors argued that reproducibility helps researchers verify findings, build on one another’s work, and move ideas into practice more reliably.

There is an element of optimism here too.

When you release code, data, models, or ideas openly, you surrender some control over what happens next.

Someone may find a mistake.

Someone may use your work differently than you intended.

Someone you have never met may take your idea somewhere you never anticipated.

There is uncertainty.

But there is also enormous upside.

Thousands of researchers can inspect, reproduce, modify, and extend something that would otherwise remain confined to a single laboratory.

In her commencement address, Joëlle described open source in essentially these terms: sharing carries risks, but inviting more people to participate dramatically expands what can be created.

The philosophy is consistent.

You accept some uncertainty because the space of possible positive outcomes becomes much larger.

The future cannot be optimized in advance

There is a temptation, especially among ambitious people, to treat life as an optimization problem.

Find the optimal degree.

The optimal lab.

The optimal company.

The optimal city.

The optimal career.

Collect enough information and, theoretically, the correct path will reveal itself.

But life is not a static optimization problem.

The environment changes.

You change.

New technologies appear.

People enter your life.

Interests you thought were permanent disappear.

Opportunities emerge that you could not possibly have anticipated five years earlier.

There is no dataset containing enough information to calculate your optimal trajectory.

At some point, you have to explore.

That may be the part of Joëlle’s speech that resonates most with me.

Optimism does not require believing that every choice will succeed.

In fact, you should expect many of your experiments not to.

Papers will be rejected.

People won’t reply.

Projects will stall.

Applications will fail.

Predictions you made confidently about your future will turn out to be completely wrong.

The optimistic belief is something different:

that acting is still worthwhile because action generates information.

You try something.

The world responds.

You update.

And then you try again.

Choose again

When faced with uncertainty, our instinct is often to wait until we feel more certain.

Sometimes that is appropriate.

But often the information we need is located on the other side of action.

Instead of trying to map your entire career from where you stand today, run the next useful experiment.

Choose an action.

Observe.

Update.

Choose again.

The future is uncertain by definition.

That is not necessarily a problem to solve.

Sometimes it is the reason to act.

Further reading

• Joëlle Pineau, “Optimism in the Face of Uncertainty” — the commencement address this post is about.
• University of Ottawa, honorary doctorate recipients
• Pineau et al., “Improving Reproducibility in Machine Learning Research”, JMLR 22(164), 2021.
• Joëlle Pineau’s website, and her research profile at McGill