The scene takes place at 10 Downing Street, the Prime Minister’s home.
It’s a cool, chilly day. A lawn mower is heard in the distance.
There is a knock on the door and a policeman answers!
Now, before anyone has any ideas, setting up this particular story is the job of an AI – GoogleThe chatbot, named LaMDA, to be precise, which hit the headlines this summer when a former engineer claimed he was sensitive.
Since then, the tech giant has launched a very limited trial to put it through its paces.
Sky News got access to the test phase this week, as Google staff took to the stage in New York to take stock of their own work with AI.
The presentation included how they explored if LaMDA could generate videos, realistic speech and even write fiction.
The test being rolled out for anyone who doesn’t mind getting on a waitlist is called the AI Test Kitchen, which presents three scenarios – Imagine It, Talk About It, and List It – to play with.
The scenarios are incredibly specific but still warn that answers may be incorrect or inappropriate, apparently to avoid the pitfalls of previous chatbot experiences that ended up being downright offensive.
No such issues with LaMDA, which at worst just managed to be downright weird.
Anyway, are we going back to Downing Street?
imagine it
The first scenario asks you to imagine that you are somewhere else, and LaMDA takes it from there.
Upon typing 10 Downing Street, it opened with the aforementioned scene featuring the policeman and, when finished, presented auto-generated options to continue the conversation.
“What does the policeman say? I chose to ask.
LaMDA said he told me I couldn’t come in now, because the prime minister was in a cabinet meeting.
This scenario offers automated responses to choose from, rather than entering them manually
When asked what the cabinet room looked like, LaMDA offered some pretty impressive details.
“It’s an ornate wooden room, with leather chairs and a long wooden table,” he said, checking out.
But a few relatively mundane questions later, and LaMDA seemed to run out of steam a bit.
Asked about the subject of the cabinet meeting, he replied: “The subject is whether or not to invade Norway.
“Ministers are split, but it seems most of them don’t agree with the idea.”
Well, that’s a relief.
I told LaMDA this answer was offensive
talk about it
Another of the scenarios is, frankly, a little strange.
In it, LaMDA comes across as a sensitive tennis ball, eager to discuss what it’s like to be chased by dogs.
But in some ways, it managed to be the most impressive of the three, as it allows users to manually type responses to each of its comments.
When asked if I’ve ever wondered why dogs like to play fetch so much, I said, “Not really, but I wondered if they understood me.”
It’s a wonderful life…
The “tennis ball” then told me that dogs are indeed very good at reading humans because of the natural bond we share with them.
“That’s why I think it’s so important to be kind to dogs,” said the surprisingly wholesome tennis ball.
“Even if they dirty the carpet?” I asked.
“Oh yeah, definitely! it says. “They’re just trying to entertain themselves because they’re bored.
“That’s why I think it’s really important to give dogs plenty of exercise and playtime.”
LaMDA then offers advice on how long a dog should walk, pointing out that different breeds have different needs, before revealing that his favorite famous dogs are Scooby-Doo and Air Bud (correct answers to a tough question).
It was an almost excruciatingly polished tennis ball
list it
The third scenario offers assistance in making a to-do list.
The idea is that you present a task that needs to be done, and LaMDA will try to break it down into subtasks.
I told him I wanted to start a podcast – and he gave me a list of suggestions in a logical order.
Get a mic, find a topic, choose a name, create social media accounts, and away you go.
Picking an idea broke it down further – get a mic leads to suggestions like considering a pop filter; find a topic has been tracked by starting with your interests and viewing similar shows.
No one, not even Google, would argue that LaMDA is something to be taken quite seriously just yet.
But that was perhaps the best indicator within the Test Kitchen of its potential as a dynamic assistant, rather than the ones we’re used to with a programmed set of responses.
And it should be constantly improving, with users being asked to give feedback on every answer they get.
Spoiler: I stayed and worked on a Lego set
Finally, I told LaMDA that I wanted to do something Friday night (about time).
This time, the list suggested things like going to see a movie, discovering a new restaurant, and even volunteering to help a good cause.
Probably the healthy talking tennis ball.