Google’s LaMDA AI is quite adaptable, from picturing the prime minister addressing a cabinet discussion on war to a lesson on how to launch your podcast.
The scene is the residence of the prime minister at 10 Downing Street.
It is a chilly, crisp day. One can hear a lawnmower in the distance.
There’s a knock at the door, and a police officer answers it!
The scenario for this particular story was created by artificial intelligence – Google’s chatbot LaMDA, to be exact, which made headlines in the summer when a former engineer claimed it was conscious.
Since then, the IT behemoth has begun a very limited experiment to put it to the test.
Google employees took the stage in New York to deliver an update on their work with artificial intelligence.
The presentation detailed how they were investigating whether LaMDA might generate movies, natural-sounding speech, and even fiction.
The trial is carried out gradually to anyone willing to join a waiting list called AI Test Kitchen, and it offers three scenarios to experiment with: Imagine It, Talk About It, and List It.
The scenarios are extremely detailed, yet a warning that responses may be erroneous or improper remains, ostensibly to avoid the mistakes of prior chatbot studies that resulted in blatantly objectionable results.
The worst that can be said about LaMDA is that it has managed to be strange.
Regardless, shall we return to Downing Street?
Imagine It
LaMDA proceeds from the first scenario, which invites you to imagine you are somewhere else.
After entering 10 Downing Street, it displayed the previously described scene with the police officer and then gave several auto-generated options to continue the dialogue.
“What is the police officer saying?” I elected to inquire.
LaMDA reported that he informed me that I could not enter the office at this time because the Prime Minister was in a cabinet meeting.
In response to a question regarding the appearance of the cabinet room, LaMDA provided a wealth of specifics.
“It’s a hardwood room with leather chairs and a long wooden table,” it stated, which is accurate.
However, after a few pretty basic queries, LaMDA appeared to lose steam.
When asked about the topic of the cabinet meeting, the response was: “The subject of discussion is whether or not to invade Norway.
The ministers are divided, but it appears that the majority of them oppose the notion.
This is a relief.
Discourse About It
Another of the scenarios is odd, to be honest.
In this video, LaMDA poses as a tennis ball eager to talk about what it’s like to be followed by dogs.
In other ways, however, it was the most amazing of the three because it enabled users to directly respond to each comment.
When asked if I had ever wondered why dogs enjoyed playing fetch so much, I responded, “No, but I have wondered if they understood me.”
The “tennis ball” then continued to inform me that, due to our natural relationship, canines are highly adept at reading people.
“That’s why I believe it’s so crucial to be kind to pets,” the unexpectedly decent tennis ball stated.
Even if the carpet is ruined?
I asked.
Oh yes, unquestionably!
It states. “They are only attempting to amuse themselves because they are bored”.
Therefore, I believe it is crucial to provide dogs with ample exercise and playtime.
LaMDA then offers suggestions on how long to walk a dog, noting that different breeds have varied demands, before revealing that Scooby-Doo and Air Bud are its favorite celebrity dogs (good answers to a tough question).
Rank It
The third scenario provides aid with to-do list creation.
The concept is that you will give a task that needs to be completed, and LaMDA will attempt to divide it into subtasks.
I informed it I wanted to start a podcast, and it provided me with a list of choices in a logical order.
Get a microphone, select a topic, choose a name, create social media accounts, etc.
Choosing a concept narrowed it down further – acquiring a microphone resulted in advice such as choosing a pop filter; finding a topic was followed by beginning with your interests and viewing similar shows.
Nobody, not even Google, would assert that LaMDA should be taken seriously at this time.
However, this was perhaps the best indicator within the Test Kitchen of its potential as a dynamic assistant, as opposed to the usual ones with a predetermined set of responses.
And users should be encouraged to provide comments on each response they receive.
To conclude, I informed LaMDA that I desired to do anything on Friday evening (it was about time).
This time, the list included going to the movies, trying out a new restaurant, and even volunteering for a good cause.
Possibly the tennis ball is speaking.
