By Jim DeCrescenzo, FAPR, RDR, CRR
There are no flying cars. Cryptocurrency has not replaced cash. Autonomous vehicles have not made driving obsolete. And artificial intelligence does not exist. Those are the facts. Well, you may be scratching your head and saying What's he talking about? I hear about AI every day. That's right, you do. But hearing about it doesn't make it a fact.
Let's begin with a definition of artificial intelligence. Research into artificial intelligence began at Dartmouth in 1956, and it encompasses the techniques used to teach computers to learn, reason, perceive, infer, communicate and make decisions similar to or better than humans. Yes, computers can store and recall massive amounts of information, far more than any human, but that's all they can do. They are limited in their capacity, just like a robot building a car is limited. The robot will do its job perfectly every time, without complaint, and never ask for a raise. But move the robot a half inch and it's useless. The same concept applies to computers. They're superfast, can store and recall massive amounts of information, but can't answer a captcha question that a five-year-old can.
So who's intelligent, the computer or the five-year-old?
The five-year-old had no language skills when he was born. In fact, he couldn't say a word until about 18 months old. And it's important to know, he wasn't taught language, since language is natural to all humans. We begin to speak spontaneously in whatever language we're exposed to. From Steven Pinker's book, Language Instinct, " studies have shown that around the age of three children are able to use the -s suffix correctly in more than 90% of the sentences that require it. For example: His tail sticks out like this. It looks like a donkey. What comes after 'C'? The children could not have been simply imitating their parents, memorizing verbs with the s's attached." A child learning a language is intelligence. No computer can do that, because words have no meaning to a computer. If you say "no" to a child, he knows you don't want him to do something because he understands your meaning behind the word.
Google and Apple use their massive storage of human speech to improve their systems' ability to respond to commands. I'm assuming this is why StoryCloud's February press release said, "the audio file is processed by thousands of computers and powerful AI software to produce a 99% accurate rough draft." I presume StoryCloud runs a draft of the transcript through, in their words, "thousands of computers" so the software can compare each word to stored recordings and generate what they say is a "99% accurate rough draft." A human with a brain still has to review each word of the transcript, because only a human brain understands context and can create a certified transcript.
I recently spoke to a person from IBM's Information and Technology Services section to talk about Watson. From my discussion with him I learned that Watson needs to be "trained" by its user "to learn the jargon" and will get better with time. My statement that Watson demos I've seen create a transcript that's 80 to 85% accurate was not refuted. The IBM representative was surprised when I told him we require a minimum of 95% accuracy for passing tests in our field. I then brought up that all litigation involves multiple speakers. His reply was, "Oh. I don't think Watson would do well with multiple speakers." In a follow-up email I told him I'd heard Watson was transcribing the captions for TV stations before the broadcast happened, and that the TV news personalities just had to read the script. I haven't yet received a reply.
Each of the speech-recognition programs is adequate for single-source commands and can transcribe audio at about 85% accuracy. But the leap from transcribing sound to real human intelligence is nowhere near. Here's an example. You're sitting at home with your family and suddenly everyone smells smoke. Even the pets will know to get out of the house. Will the computer have even a pet's intelligence to escape? No. The humans and pets will escape because they've gone through an evolutionary process that gave them intelligence, albeit different levels. Computers have zero intelligence.
Technology is wonderful. We can pick our phones up and call anywhere, search for anything, and do countless other things. This is current technology. However, future technology, as presented by developers and the media, is often terribly inadequate, and often never gets beyond the hype stage and into practical use. (Think cryptocurrency, autonomous vehicles, flying cars.) When you talk about screen doors and suddenly ads appear for screen doors, that's not artificial intelligence. It's no intelligence at all. The computer is simply using its massive storage and recall capabilities to match the sound of your spoken words to its audio library, and up comes the ad. It displays the ad because programmers told it to.
There is no intelligence or understanding by computers. It's simply doing what they do best, storing and recalling.
Transcribing multiple-speaker human speech is the holy grail of understanding. We are able to do this, often with minimal effort, because our brains understand speech in context. If I say, "Kilauea Volcano is becoming very active," you immediately understand there's a risk of eruption and all the consequences that will have. A computer understands nothing; it simply does what? That's right, it stores and retrieves data. It has no intelligence.
This is a massive topic that really needs a full seminar to cover. Let me end by saying I am certain that this great profession can have a bright future, if there are enough of us to do the job. AI is not the threat. The real threat is our failure to replenish our ranks.