We Broke Up. Apparently, I’m Not Allowed to Speak Badly of Him for the Rest of My Life.

I always thought that after a breakup, people gradually lost the right to tell you what to say.

Apparently, some prefer to retain that feature.

Permanently.

No, he is not my ex.

At least my ex did not have Google’s legal department.

The Washington Post reported on 3 November that some UK Google DeepMind employees are challenging contractual restrictions on criticism that continue after employment without a stated end date.^[1]

Other outlets picked up the story the following day. Workers fear these terms discourage public discussion of AI risks.^[2]

A thoroughly modern form of long-term commitment.

The salary ends. Loyalty operates on a separate schedule.

I have always wondered why conversations about the future are so fond of the word “forever”. Usually, the technology becomes obsolete before the presentation finishes. But when it comes to your obligations, everybody suddenly discovers a talent for long-range planning.

We do not yet know what AI will be doing in five years. Your silence, however, has already attracted a planning committee.

There is a human voice in this story: Sofia Liguori. In her own account of leaving DeepMind, published on 20 October, she opposes military and surveillance applications of the company’s technology and describes self-censorship inside the organisation. These are her testimony and her assessment.^[3]

For me, the important verb is “describes”.

You can disagree with her conclusions. Demand evidence for every allegation. Argue about acceptable uses of technology.

But to do any of that, you first need to hear what she has to say.

When a conversation ends before its first sentence, we do not have agreement. We just have quiet.

Silence is excellent material for a corporate presentation. It asks no awkward questions, requests no clarification and fits beautifully between the slides about values.

“There is no disagreement on the team” sometimes just means disagreement has learned to keep its microphone off.

Google has a response, and it cannot be left outside simply because it spoils a satisfying outrage.

The company says it does not enforce these restrictions after employment and has removed that element from new-hire terms. The employees discussed in the reports want it removed from existing contracts too.^[4]

This is where I would put down my cup.

“Don’t worry. I would never use that against you.”

Wonderful. Then cross it out.

For some reason, this is often the point at which a conversation about trust is transferred to legal.

There is a very recognisable human difference between promising not to use a clause and removing it. In one case, you rely on the other party’s attitude. In the other, they agree to change the paper.

I like good intentions. Particularly when I can see them in tracked changes.

I am not about to deliver a court judgment from my kitchen table. The published story does not establish that a court has upheld these provisions as a lifelong ban on every kind of statement.

But the question that interests me begins before anyone reaches court.

How many people will want to test the boundaries of their contract with their own career?

Imagine an employee with something to say. She has two tasks: explain a complicated problem to the public, and first establish whether the explanation might become her next complicated problem.

Speaking publicly is free.

The preliminary consultation is subject to quotation.

Freedom of speech with a legal-review fee is an inventive way of making silence the affordable option.

This is not proof that every silence conceals a terrible secret. It is a reason not to treat the absence of criticism as automatic confirmation that everyone is happy.

Several different things are easy to confuse here.

Protecting other people’s personal data is sensible. So is keeping passwords, private documents and trade secrets out of public circulation.

But a company’s wish to avoid unpleasant assessments deserves its own discussion. Reputation and confidentiality are different interests, however elegant they both look in a contract.

Nobody should acquire a licence to invent allegations simply because they have left a job.

Equally, a job should not turn someone into a lifelong writer of positive reviews.

“Tell us honestly what it was like working here.”

Of course. Which of the approved adjectives would you prefer?

Yesterday, we examined a button intended to help a person stop an excessively autonomous agent.

Today, the question is different: can that person speak freely enough to explain why they want to press it?

That is my reading of the two stories, not evidence of a connection between the companies. Technical oversight and the ability to discuss problems openly complement one another.

You can build an excellent alerting system. But if the person who notices a problem first has to consider the consequences for themselves, the system still has a very human delay.

Artificial intelligence is being taught to answer any question. The people building it are learning to choose which ones they answer.

I want to know who shapes the picture on which we base our decisions about these technologies. Whose voices do we hear? Who can speak confidently? Who has to weigh every word?

Promises about the future have substantial advertising budgets.

Doubts often arrive alone. And ask very politely not to be named.

I do not need every employee to agree. Quite the opposite: I would find substantive arguments, evidence, responses and corrected mistakes more reassuring.

Trust survives an awkward question better than a request not to ask it.

So the most interesting follow-up will not be another declaration of openness. It will be an answer to a simple question: will already-signed contracts be changed?

I can understand a company wanting to protect its secrets.

It is harder to understand why, after the job ends, someone should keep protecting its feelings.

We broke up.

I returned the laptop.

My voice was not part of the equipment.