I’ve been getting notes from all corners the last few days about a $200 virtual DJ program that has been been given her own show on radio station KROV in San Antonio. The program is actually an application of a more general “virtual assistant” product from Guile 3D Studios. DJ Denise goes on the air at KROV tomorrow, from 1 PM to 4 PM. You can listen over the Web; I intend to.
Most of the horsepower in creating Denise seems to have gone into rendering her lip gloss, which is odd for something used as a broadcast audio DJ. I’m more interested in whatever AI lies behind the pouty face, though early indications are that she has more lip gloss than AI. It’s an issue of special interest now, because I’m making slow but steady progress on a near-future SF novel that explores (among other and stranger things) the border between real AI and “fake” AI, a category that goes back to the ELIZA program at MIT in 1966. Ten Gentle Opportunities contains both. A cheap coffee maker contains an animated barista that talks a lot and understands little. A model-year 2020 Mazda RX9 has an annoying dashboard cartoon that understands more deeply but very narrowly. Both handle natural language well, and people are easily fooled into thinking that smooth natural language processing implies true intelligence.
I don’t think that’s true, as the more advanced AIs in the story demonstrate. One of them is Pyxis (Latin for “compass”) a high-end commercial product sold at a five-figure price as an executive assistant. Brandon Romero, an executive trying to manage a completely automated AI-controlled copier factory, has his own copy of Pyxis. (I posted a glimpse of the copier factory and its AI controller Simple Simon in my June 26, 2011 entry.) Far from being a geek-dream sex kitten, Pyxis is obedient without being especially pleasant. Worse, she holds her boss to his word, to the point where he begins to wonder who’s working for whom.
Romero dislikes having human underlings, but as he soon comes to understand, AI staff might be described the way Jerry Pournelle once described the Bomarc as the Civil Service missile: “They don’t work, and you can’t fire them.”
Pyxis saw him approaching his office door, and Brandon heard the lock bolts snap back. The coffee machine on the teak credenza was hot and full, and the air was rich with the scent of dark roast and Irish Crème. The human interns always scattered magazines on the glass coffee table against his preferences; the day when paper magazines became extinct could not come too soon. One of those interns had recently left a stuffed moose on the credenza. This was at the direction of HR, which (as he later discovered after much annoyance) wanted to “soften the human side of his persona.” The ugly abstract art shotgunned at the eggshell walls was bad enough. God forbid he should meet with a Chinese parts supplier without his stuffed moose.
Brandon sat down at his teak desk, its oiled vastness divided into the rigorously rectangular regions he maintained at all times, including a small square for coffee and another for mints: charts, summaries, two tappers full of notes and test-run videos and model animations, all at his fingertips. Defining the far sides of his desk were three brushed-stainless OLED panels currently animated with some slow-flowing pearlescent liquid that looked like shampoo. Far too soon, the triptych would spring to life with more views of this lunatic’s kingdom than any one man could possible follow.
Pyxis saw him sit down, and a window in the panel to his right burst into existence with her scowling image. “Twenty-six messages vetted and queued, five urgent.”
“Later.” If it wasn’t from that ass-covering coward Amirault, he didn’t want to hear it. Brandon set his primary tapper down in its vacant rectangle on the desk, and pulled a few loose papers from his briefcase. Like everything else, each had an appropriate place, and he scanned the piles that had been accumulating for most of a week, dropping a sheet here and a sheet there. The stapled set describing Zircon’s looming Retirement Incentive Program (was that a hint?) needed to go somewhere. A new pile? For corporate suicide notes? Brandon scanned the desktop almost automatically, but there was only one empty rectangular region left.
He stared at the tidy strip of oiled teak and felt himself tighten inside. Not big enough for anything except bad memories-but like those infuriating little sliding-square plastic puzzles, he had never hit upon an arrangement that would eliminate it.
“Here it is, Mr. Romero.” A new window popped into view, with a high-res scan of the framed photo that had stood in that teak rectangle for many years: Carolyn in a white cotton V-neck sundress out in her garden, holding a cardboard sign reading, “Greek Fire.” To a newly minted second lieutenant on the ground after Desert Storm, it meant that Carolyn Helena Ankoris was waiting impatiently for him to come home and marry her. To Major Brandon Louis Romero, US Army, Retired, it meant only failure.
“I didn’t ask you to open that.”
“You were staring at the space where the photo had been.” Building 800 was as full of electronic eyes as it was empty of human beings. Pyxis not only knew where he was at all times, she knew where he was looking.
His AI assistant was unfailingly obedient, but Brandon had set her obsequiousness parameter to zero. What was the point of having a virtual suckup? It wasn’t like the physical world suffered a flunkie shortage. “Your job isn’t to read my mind.”
Pyxis folded her arms implacably. “My job is to anticipate your needs and help you stay productive. We have a line start in a little over an hour. You have a lot to do. Mr. Amirault asked you to copy him on a call to…”
“Ok.” Brandon tossed back the last of his Red Hen coffee, and flashed with sad longing to his Army B4 training, when he had aimed an M16A4 at line-drawn enemies printed on sheets of cardboard, and nailed every damned one through the heart. “Get me Simple Simon.”
Have you seen the Red Dwarf episode with the talking toaster with AI? It’s hilarious. The toaster keeps pestering them to have more toast, crumpets, muffins, bagels, anything. It gets depressed if they do not take its offers. “I think therefore I toast.”
I’ve never seen an episode of Red Dwarf and I’m not sure where it ran in the US, though if I see it on somewhere I’ll take a look. My AI coffee machine in the novel keeps pestering its owners to upgrade it, but it’s a minor character at best, appearing mostly in one scene.
“Denise” probably tells us something about the amount of intelligence radio stations expect from their “talent.”
The lip gloss and the Star-Trek-computer voice may be the major appeals to a certain segment of young male nerds. This may even be the radio program’s target audience. I can’t tell as the listen-over-the-web link doesn’t work for me.
Short term, the buzz from being the first robot DJ will increase market share. Long term, who knows. The vendor promises lots of improvement. (Meaning, they know there’s plenty of room for improvement.) Maybe, the pig will learn to sing.
Well, the AI DJ is a real disappointment. The guy doing the experiment is just plain WRONG that this has not been done before. *I* have done it, and so have others.
Clearly the voice is the Nuance text to speech voice, Samantha–the same voice Kindle uses on its speech function, and also the one that Extra Normal uses for its online teddy bear dialog machine. And if the guy spent $200, that is $80 more than he would have had to spend, because you can create Samantha and Tom’s voices using a Kindle, and just importing a text file for it to read, then recording the output.
On the West Coast, some radio stations there have been using Apple speech synthesis to announce song title and artists, in the same manner that newer iPods will do while they work their way through your favorite playlist.
I am involved with a community radio station in a nearby town, and I have created bits to play on the radio using both Samantha and Tom’s voices. Others around the country have done similar things; perhaps it is true that no one has ever used the voices to entirely substitute for a DJ,–but even that I find unlikely. It has definitely been used to create weather forecasts and time/temperature readings for stations to play, in many places around the world.
Meanwhile, there are others who are doing much better text to speech than Nuance. Acapella is just one. You can try out their voice, Heather, online, and it sounds a whale of a lot better than Samantha–even after KROV changed her name to Denise.
And in one article I read, the guy whose experiment this is, claims that you can get someone to program the voice for $10/hr. This is typical of small market thinking that does not understand creating entertainment for the masses (although San Antonio is not a small market). When Grant Tinker (Mary Tyler Moore’s husband) left his wife’s company to become head of NBC, he was asked whether or not all of NBC’s programs would now be as good as what came out of MTM productions (Mary Tyler Moore Show, WKRP in Cincinnati, Hill Street Blues, Remington Steele, Newhart, Lou Grant). His response was that there are not enough creative people in the country to make that many good shows–and besides, most of them were at MTM.
He is right. Program directors in small market radio very often have no understanding whatever of what it takes to satisfy audiences creatively, until they try their hand at large markets, where they fall flat on their faces with their provincial views, and disappear quickly. I assure you that $10/hr does not even come close to buying a copy writer of news or commercials in the city where I now work, Indianapolis–much less someone creating copy for program hosts. All I can say is ‘luck to you, Dominique.’ This was another case where the over-hype made for an anticlimactic outcome.
My reaction entirely: This was a publicity stunt. There is no way I would listen to that voice on a regular basis and enjoy it.
I have an unfinished short story about a small-town 1-man radio station that buys an AI DJ system and discovers that there is a sort of “uncanny valley” in audio as well as visuals. I need to finish the damned thing; it’s been sitting on disk since the late 1980s. If I decide to go back to it, I hope I can pick your brain about small radio-station tech.