Scientists Develop AI That Can Turn Brain Activity Into Text 30
An anonymous reader quotes a report from The Guardian: Writing in the journal Nature Neuroscience, [researchers from the University of California, San Francisco] reveal how they developed their system by recruiting four participants who had electrode arrays implanted in their brain to monitor epileptic seizures. These participants were asked to read aloud from 50 set sentences multiple times, including "Tina Turner is a pop singer," and "Those thieves stole 30 jewels." The team tracked their neural activity while they were speaking. This data was then fed into a machine-learning algorithm, a type of artificial intelligence system that converted the brain activity data for each spoken sentence into a string of numbers.
To make sure the numbers related only to aspects of speech, the system compared sounds predicted from small chunks of the brain activity data with actual recorded audio. The string of numbers was then fed into a second part of the system which converted it into a sequence of words. At first the system spat out nonsense sentences. But as the system compared each sequence of words with the sentences that were actually read aloud it improved, learning how the string of numbers related to words, and which words tend to follow each other. The team then tested the system, generating written text just from brain activity during speech. The system was not perfect, but for one participant just 3% of each sentence on average needed correcting -- "higher than the word error rate of 5% for professional human transcribers," the report says. "But, the team stress, unlike the latter, the algorithm only handles a small number of sentences."
"The team also found that training the algorithm on one participant's data meant less training data was needed from the final user -- something that could make training less onerous for patients."
To make sure the numbers related only to aspects of speech, the system compared sounds predicted from small chunks of the brain activity data with actual recorded audio. The string of numbers was then fed into a second part of the system which converted it into a sequence of words. At first the system spat out nonsense sentences. But as the system compared each sequence of words with the sentences that were actually read aloud it improved, learning how the string of numbers related to words, and which words tend to follow each other. The team then tested the system, generating written text just from brain activity during speech. The system was not perfect, but for one participant just 3% of each sentence on average needed correcting -- "higher than the word error rate of 5% for professional human transcribers," the report says. "But, the team stress, unlike the latter, the algorithm only handles a small number of sentences."
"The team also found that training the algorithm on one participant's data meant less training data was needed from the final user -- something that could make training less onerous for patients."
It's time (Score:1)
Long time coming (Score:3)
Books have been doing the opposite for thousands of years.
Maybe not too surprising (Score:2)
People can lip read. therefore if you just measured the neural signals controlling the lips and jaw a computer should be able to lip read even better than humans can from the visual input.
thus you don't need to read someone's mind to extract this information. the Jaw and lip motions are not so much mind reading as the translations of thoughts into signals to actuators. That is it is no more mind reading than seeing me pick up the salt shaker. You can infer from my movements that I was probably thinking
Re: (Score:2)
I blame the output on a bad spell checker.
Error rate (Score:5, Interesting)
Yet the police routinely transcribe interviews and then destroy the recording. The detective that Prof. James Duane brought into his lecture said exactly that. The transcription is accepted as 100% correct.
Re: (Score:1)
Re: (Score:2)
Yet the police routinely transcribe interviews and then destroy the recording. The detective that Prof. James Duane brought into his lecture said exactly that. The transcription is accepted as 100% correct.
What he said is that it's the witness testimony from the cop that interviewed you that is the evidence. Neither a recording nor the notes are required, that's just extra. Note that this was about the cop's own notes, if you've signed a statement like a confession that's different. Duane even said as much in one of his tirades, one reason not to talk to the police is "It can hurt if the police officer is transferred to Minnesota, or deceased, or injured, or comatose, or can not be located at the time of the
Re: (Score:2)
I sit corrected. So the cop witness testimony is de facto accepted as a 100% accurate recollection... and the raw material that may provide nuance, context, or even disagreement with the witness testimony is destroyed.
So tell me why (Score:3)
Researchers from the University of California, San Francisco publish a document on their publicly funded research, but it's archived where the tax paying public must pay or send email with a "pretty please" content.
Guaranteed the cost one has to pay doesn't go to the research.
Who here has liberated the publication via an accessible route?
Computers learning, or people learning? (Score:3)
At first the system spat out nonsense sentences. But as the system compared each sequence of words with the sentences that were actually read aloud it improved, learning how the string of numbers related to words, and which words tend to follow each other.
Or maybe rather than the system, the neural networks inside the brains of the participants were learning and adapting, leading to an increase in accuracy.
What? People can learn and improve over time? Crazy talk! If that were true, then any static electrical signal-to-text algorithm could be used (as long as it can be consciously modified), with the added benefit of it not needing to adapt to anyone.
How about those who don't think in words (Score:3)
I normally don't think in terms of words, but with abstract concepts when I need to communicate them, I need to translate what I am thinking to words.
Hopefully, it will pick up words you are about to say to use vs the train of thought. Because if it is a woman who is checking to see if this works, My sentence, depending how she looks, may be broken up with "look at the eyes, not the chest", then will be broken up with "I have been looking at her eyes too long, I think I am starting to get creepy, look somewhere else"...
No matter how cool it sounds listening to a person's thoughts will not be too useful, as it is a conflict between instinctive nature, your own self-ego, your rational planning, forecasts of outcomes on what people may say and responses to it. There is just a lot going on behind a normal point of communication. And I am sure your personal ego would get hit hard when someone thinks about you, even if they do like you.
Let's try it (Score:2)
I'm concentrating, I'm seeing an adult male of the species Bos taurus sitting on a toilet.
Must think in Russian (Score:3, Funny)
Re: (Score:3)
Re: Must think in Russian (Score:1)
Well (Score:3)
Psi Cops are scanning me! (Score:2)
lets start by correcting baseline errors. (Score:2)
a few things to correct in the thinking:
There is no such thing as artificial intelligence (cept by the brains of those promoting their fantasy beliefs as being reality) what it is actually is abstraction processing. And in this case they are using what is being called Machine learning in training what is called Deep learning or more accurately using a mathematically abstract weighted system to train abstract pattern matching.
early voice to text dictation system such as Dragon Systems https://en.wikipedia.o [wikipedia.org]
Re: (Score:2)
Sorry, but Machine learning is a form of artificial intelligence. you don't get to define artificial intelligence as "Thinks like a human" just because you don't like the current definition of the word.
I can hear the facial bones starting to crack... (Score:2)
"Full stop!" the captain shouted, and the engines ceased turning.
It is imperative we do not let governments argue brain reading is physical evidence, and so may be gathered with a warrant. No, that would violate the principle that one cannot be forced to testify against oneself. So not even with a warrant.
This is a science-fiction level dictator's power, the ability to check on thought crimes. With eyes in the sky recording all public movement every 1 second, and social good citizen scores ala Black Mirr
This will likely confirm what most women suspect (Score:2)
Are we SURE we want a transcript of what men are thinking?
https://www.telegraph.co.uk/me... [telegraph.co.uk]
I mean granted, it's text not images so we won't have to sort all THAT out, but I think you get my point.
Do we just run this, and then run a macro to filter out all sentences with references to genitalia or sex, to come up with the... 12%? ... of men's thought activity that's meaningful?
Re: (Score:2)
I came here to say this. On the other hand this is an amazing idea because I have no idea what my idling brain thinks about - it would be fascinating to know even if it is just "is it time to eat yet?" for about 80% of the time. lol.
The obligatory... (Score:1)