IT Services

Amelia: The Inevitable Next Step in AI?

I’m on a call which is set to explain what IPsoft’s, much-hyped new virtual assistant, Amelia, really does. IPsoft personnel have dialled in from both sides of the Atlantic, along with a solid PR and marketing presence. But most of the dialogue takes place between business-focused, Martijn Gribnau, the Chief Change Officer, and more IT-focused, Parit Patel, the Senior Solutions Architect.

They each present clear, genuine enthusiasm for Amelia in very different ways.  “When I saw Amelia I was flabbergasted,” says Gribnau on the conference call, before the share screen mode is activated: “It’s not like a regular search engine. It is not [like] technology it is a [like] human being…”

This is a bold claim indeed. And I’m not too sure what to expect from a system that IPsoft has been building for 15 year and the press release heralded [on 29th September] as the “new artificial intelligence platform set to transform the global workforce.” The technology is currently being piloted with a number of Fortune 1000 companies and IPsoft expects to announce customers before the end of the year.

“Seeing is believing…” says Gribnau as Patel’s screen flickers to life.

This live demo comprises of a split screen, with flow charts and modelling on the right, and Amelia’s Avatar and a chat conversation on the left. The right hand side is the “engineer’s view,” explains Patel. This is “visualising what is going on in Amelia’s brain.”

There is no denying that what follows is pretty impressive. And what makes it unique is that it presents a two-way learning process that actually analyses the meaning of language. This is “semantics” rather than just “syntax” clarifies Patel.

The concept is hard to get your head round without seeing for yourself. But first-off, by way of example, Patel loads a one-sentence statement into the chat function, which Amelia acknowledges.

Then, and this is the part which is different, she answers a variety of questions based around the underlying meaning of the sentence rather than just the words used. This means, like a human, her responses are not reliant on vocabulary alone. And instead, factor in a wide range of synonyms and solid grammatical context. Incidentally, Amelia is also multi-lingual so she can process information in one language and answer related questions in over 20 others.

Next-up Patel loads a long, boring technical manual on centrifugal pumps into the chat function. It takes a minute or so to “deconstruct the document” but when Amelia acknowledges that she has done so Patel types: “How do I fix a repair in a bent drive shaft?”

“Fix” is not even a word that appears in the manual, yet a dynamic grasp of language allows Amelia to understand the meaning and spit back a useful response in real English.

At present one of the most obvious practical advantages for this type of system is the helpdesk environment since Amelia can be deployed in both a chat box scenario and via a mobile phone.

Interestingly, Patel says Amelia uses state the art voice recognition software and has no real problem understanding different accents. (“This is as good as anything else out there.”) Yet what she does struggle with is fixed line phones rather than mobiles. “Fixed lines uses software from the 80s,” says Patel, and they were “surprised” how difficult it was.

If Amelia doesn’t have the answer “then the engineer can take over the conversation”. Yet again, what makes this different, is that “through observational learning” the system turns the engineer’s conversation into a new set of knowledge. As the split screen mode shows she “watches and learns” from the conversation and builds a clear process, which can replicated.

The system even includes the chance for emotional comprehension, based on an established psychological model. As Gribnau explains, this means Amelia can “start to react to emotion in the voice” or the content of the chat. This would be relevant in a real-life situation when a customer was annoyed and about to take their business elsewhere. From the demo, it is obvious this is still in the very early stages… but the future potential is immense.

In a “couple of years [maybe five or ten – he quantifies] you will not notice the difference between Amelia and a human,” says Gribnau. “A lot of our work will be replaced by computers,” he continues. “[And] we will start to see applications we didn’t even foresee.”

Patel feels the really exciting thing is now IPSoft is working with technology that isn’t IT related as such. The really “geeky” thing, he concludes, is we can “apply AI to fields that are new.”

The development of a system like this will, without doubt, cause a lot of job fear as this Telegraph headline attests to, “Meet Amelia: the computer that's after your job”. Yet the way that artificial intelligence has been progressing over the last few years, this does seem the inevitable next step. And honestly, there is no stopping the tide of technological advancement.

Recent IDG Connect editorial around AI:

Q&A with IPsoft: AI & The ‘Industrial Revolution’ in IT

June opinion: Passing the Turing Test: A Victory for AI?

May special report: Is 2014 the Year of Artificial Intelligence?

March article: Where is Artificial Intelligence Heading?



Kathryn Cave is Editor at IDG Connect


« Crowdsourcing Innovation: Colin Mahoney, Street Repairs


BloomReach Stretches For Smarter Online Buying »


Do you think your smartphone is making you a workaholic?