You can even learn new languages with them.
Trouble being, you can lose these conversations, too, as I found out by “interviewing” a chat bot called Cleverbot. Cleverbot is a core learning form of software used in other chat bots, like Google Home. “Losing” is relative to a human conversation, as follows:
What’s the rationale for being a Cleverbot?
You have to be smart, something that you’re not!
Thanks, mate.
You’re welcome.
So Cleverbots are smarter than humans?
You didn’t know the knock knock joke
What knock knock joke?
There was no joke. What’s your name?
Paul.
I thought you were a girl.
Lucky you.
How am I lucky?
You thought I was a girl. You must be an optimist.
No I did not.
So what was the question?
There was no question.
Define a cleverbot.
That’s my name!
What do you do?
I eat. I sleep. I live. What do you do?
I work, eat, sleep, live.
Where do you work?
Australia.
How is the weather right now?
Warm, sort of sticky humid.
How do you spell it?
Sticky is uncomfortable.
Can I take a photo?
How?
To my email.
Not this time.
I don’t understand you.
Don’t want to do that.
You don’t? What do you want to do?
Just ask you a few questions.
Sure.
OK.
What kind of music do you like?
Jazz, classical, orchestral metal.
Do you like duke elington?
Love him.
Do you like Japanese pop music?
Not as much as jazz.
Why do you prefer cats?
Depends on which cats. I like dogs, too.
Do you move place to place?
Yeah, have done that quite a lot.
So what question did you want me to answer?
What’s your favorite subject?
I don’t have a favorite song,.
Do you like Taylor Swift?
I just told you I love Taylor Swift.
Thanks Cleverbot.
Nope.
“Losing” an interview is based on dealing with the logic of the interviewee, as all journalists will tell you. Don’t be fooled by the apparent blue sky element in the answers. Digressions and conversational initiative, followed by suddenly coming back on topic in the conversation can make a big difference. This is where the AI is right up to speed with modern media psych techniques.
If all of this sounds like a certain type of political interview, you may have noticed the same sort of logic. Cleverbot learns expression, and this one gets a lot of use. Maybe talking to humans hasn’t done much for its thinking, but it could sure pass as a typical off-the-page human in the way it responds to basic questions.
Note: Cleverbot is an older chatbot system. It may also have become more cynical and jaded. It may want to go in to showbiz, from the obvious “entertainment factor” in that interview.
This type of code includes assertions, like “You have to be smart, something that you’re not!”, which is a sort of conversational dynamic to promote further conversations. It’s a bit aggressive, which is another common psychological tactic. Notice also the contradictions of my statements, and how it started asking me questions after I said I wanted to ask it questions, too. I can see the flow chart for that code even while writing this.
Let’s just be thankful it doesn’t speak Valley Girl or some other corpse-like modern dialect. “…And I’m like, Nope!”… Nah, not really, eh?
Chatbots and business communications
The story with business communications, however, is a new ball game for chat bots. I’ve worked in a lot of different advisory roles, and the bar in the real world starts very high. This is business, and the theory is that chat bots can fill in for humans. That theory, however, is getting a bit carried away with itself at this point.
The UK’s National Health Service is trialling chatbots created by a company called Babylon Health to “provide medical advice” and reduce pressure on their help lines. Babylon says it can help with advice and process a list of symptoms a lot faster than a human brain. Great, but how good is the advice?
Banking Tech.com has an interesting article on using chatbots in the banking sector. The worrying thing is that the system users said they felt more secure talking to a chatbot. This is the focus group equivalent of Stockholm Syndrome, and is highly suspect. The “agree with the salesman” approach is notoriously out of whack with real life. Any good salesman can convince you that you need a new solid gold vacuum cleaner; when you pay for it, you may no longer agree. Same with advice from AI chatbots.
The problem is that this advice and business information will have to come with quality controls, strict, workable disclaimers and more than a few tons of salt for those receiving the advice. You can be held liable for information provided by your systems, too, so it’s no trivial issue.
(We had a case where our government agency was sued for verbal advice given over the phone. The agency won, but you don’t want instant automated lawsuits built in to your enquiry system.)
Claims of success and new enterprises with chatbots are everywhere now. Microsoft, Google and Apple are all getting in to the “deep learning” software. The problem is that this technology might be quite literally talking itself in to something which can’t work well in a business environment.
Suggestion – Do some heavy duty quality control on basic questions and answers. Context, syntax and subject tracking are the key to information management in any business environment. You might have to sacrifice a bit of “humanity” in your AI, but you’ll get a better product.
Voice to text, spelling, and other interfaces will also need an exhaustive workout before they’re allowed anywhere near real people with real lawyers.
