Blog Post

The next step for intelligent virtual assistants: It’s time to consolidate

Stay on Top of Enterprise Technology Trends

Get updates impacting your industry from our GigaOm Research Community
Join the Community!

When we talk about the future of artificial intelligence (AI), the discussion often focuses on the advancements and capabilities of the technology, or even the risks and opportunities inherent in the potential cultural implications. What we frequently overlook, however, is the future of AI as a business.

[company]IBM[/company] Watson’s recent acquisition and deployment of [company]Cognea[/company] signals an important shift in the AI and intelligent virtual assistant (IVA) market, and offers an indication of both of the potentials of AI as a business and the areas where the market still needs development.

The AI business is about to be transformed by consolidation. Consolidation carries real risks, but it is generally a sign of technological maturation. And it’s about time, as AI is no longer simply a side project, or an R&D euphemism. AI is finally center stage.

IBM, for all its investment in the Watson platform, was still missing, among other elements, the “personality” — a critical piece of the virtual assistant puzzle. IBM is betting big on Watson overall, to the tune of $1 billion, and is therefore addressing Watson’s weaknesses aggressively. Assembling the complete puzzle is a non-trivial technological challenge and I’m not at all surprised IBM snapped up Cognea.

IBM Watson. Photo by Clockready/Wikimedia Commons
IBM Watson. Photo by Clockready/Wikimedia Commons

One of the companies I advise, [company]Next IT[/company], has gone to great lengths to create IVA’s with fully-developed personas — ranging from SGT STAR for the [company]U.S. Army[/company] to [company]Aetna[/company]’s Ann. These IVAs have a tone, a personality, a sense of humor and a vernacular custom-suited to their use cases. But even those personalities required proficiency in other facets of the technology, such as an expertly developed domain model. If you chat with the above examples, you’ll see how the two pieces interact very differently.

Because intelligent virtual assistants are focused within a domain model, they benefit from a clearly defined knowledge base and are able to go much deeper and stay within those bounds, whereas general purpose assistants like [company]Apple[/company]’s Siri are often asked to deal with users’ wide-ranging and often disorganized goals.

This is yet another argument for consolidation — building and implementing a cross-domain view of the world is a challenge, very likely bigger than any single company or customer. We likely won’t have “one assistant to rule them all,” but rather a team of assistants, each aware of its strengths and weaknesses, always collaborating in the background, delegating and stratifying based on the task at hand.

Personality is just one example of a shortcoming being addressed. So-called “best of breed” capabilities are scattered all over, and right now the AI market has a lot of niches. Natural language processing is important; machine learning — both statistical and symbolic — is important; domain models and ontologies are important; reasoning is important. The list goes on.

Cortana, Microsoft's virtual assistant. Photo by Kevin Tofel/Gigaom
Cortana, Microsoft’s virtual assistant. Photo by Kevin Tofel/Gigaom

These disparate focuses inherently mean most IVAs are incomplete — they’ve either gone to college and have weak people skills, or they skipped college altogether and have some street smarts but little else. They either partied too much or spent too much time in the lab. To echo my point above, they’re not well-rounded.

So consolidation has become the necessary next step in the market. Or, to mark this drive to consolidation with a phrase most technology executives know all too well in accordance with Gartner’s Magic Quadrant, we are now beginning to realize “Completeness of Vision” for AI.

Personalization may be the biggest impediment to achieving that completeness of vision right now. Siri, as an example, is fairly capable as long as you know when to use her. But your Siri is identical to my Siri, despite various claims to the contrary. Because of that, our relationships with her are shallow. Siri’s personality is one-size-fits-all, a quality that prevents greater intimacy.

Early reviews suggest [company]Microsoft[/company]’s Cortana might have a fighting chance, at least in terms of the raw number of tasks she can carry out on your behalf. Cortana is a great multi-tasker, but still misses any notion of adapting to the individual.

[company]Nuance[/company]’s Nina, on the other hand, is strong on natural language recognition (as is Cognition), but weak on real intelligence (learning and reasoning). Other players have a lot of brains under the hood but lack real use cases. And it will be interesting to watch new entrant Viv develop and see what areas it focuses on and what it lacks.

Cognition as a service: The next OS battlefield

Some companies are still only using generic natural language processing and some clever front-end work to create the illusion of intelligence, but have managed to gain an impressive number of users by focusing on user experience. Other new startups have focused on applying these smarts to everyday objects — AI is a natural friend of the IOT (internet of things), and will be key to its development. Even [company]Yahoo[/company] is trying to get off the sidelines and into the game.

Photo by Hadrian/Shutterstock.
Photo by Hadrian/Shutterstock.

There’s a lot of fresh capital flying around in the AI world generally. Industry analysts are already predicting massive growth in the virtual assistant market — as much as 39 percent CAGR between now and 2018. Although the market is vibrant, no one has yet managed to string together the holistic experience that will be necessary for the future of AI as a business.

Consolidation will finally bring about the realization of an end-to-end “cognition as a service,” and as I’ve argued before, CaaS is the next OS battlefield. This is platform wars 3.0. The big internet companies are already spending a lot of money in this area, and they’re only going to spend more.

A real end-to-end platform would make it possible to completely reinvent how we interact with machines in health care, customer service, travel, finance, commerce and more. But the market needs consolidation to achieve that potential, because the technologies at play are unusually complex, and regrettably siloed.

Whether that consolidation comes through M&A or meaningfully executed strategic partnerships, or through standards development and the open sourcing of key technologies, we’ve reached a point of adequacy with AI. More collaboration, cross-pollination, and integration is necessary in order to take the next step.


Nova Spivack is co-founder and CEO of real-time trend dectection company Bottlenose and an advisor to Next IT. His background in AI includes time at Individual Inc. working on AI-based news filtering, and at Ray Kurzweil’s neural network company, Xerox/Kurzweil, and Danny Hillis’ supercomputing venture, Thinking Machines, as well as research at SRI as part of the CALO project (which became Siri), and the founding of Twine, a pioneering B2C semantic web startup. He previously wrote about consolidation of the quantified self market for Gigaom.

8 Responses to “The next step for intelligent virtual assistants: It’s time to consolidate”

  1. Peter Fretty

    It’s truly the convergence of a host of disruptive technologies that is making this ongoing evolution become a reality. When cloud computing, mobility, and big data meld together it presents an opportunity for seamless interactions. Of course, the key here is that this is still an evolution. Most of us have not seen a true virtual assistant in action. This requires access to a healthy data base filled with clean quality data. It also means the infrastructure is in place. Unfortunately most organizations are still struggling to handle many of the pivotal tasks that make big data successful – as outlined in a recent SAS survey. However, the fact that we are looking forward, establishing goals and evolving shows that the needed level of maturity may be in sight.

    Peter Fretty

  2. Microsoft’s Cortana is taking a stab at consolidating specialized personal assistants. For example, when Cortana is active, you can simply say, “Insteon, raise living room temperature” to control Insteon’s home control system, and, in the next breath, say, “Cortana, find a good Indian restaurant nearby.” They’ve built support for this feature into their tools. This functionality will evolve into, in effect, a search engine for specialized apps, and will be emulated by other general personal assistants. I’ve argued in several forums that every company will need a specialized personal assistant, just as they are expected to have a web site. As Spivak noted, these specialized assistants can support more complex dialog because of the more limited context.

  3. Thanks for the heads up Steve on G+ and greetings Nova. In upper portion of article I wasn’t sure whether you were calling for market consolidation or technical consolidation–both are occurring of course, but then in final paragraph you were clear on being uncertain. I can share that I worked at the board level with several of the market leaders for years from a flexible perspective towards functionality & can report primarily CI. And that was with decades of relationships–some strong. If we embrace standards as my company does, and SW engines provide functional APIs, then customers can bolt together whatever they need. Our architecture encourages it–we provide the enterprise governance, IP, and platform, a few essential applications and functions, but beyond that we encourage customers to the point of making recommendations when aware to seek out internal and third party algorithms and applications–despite the high level architecture, and deep functionality, or slice is quite small as can be observed in our published price list. But of course we also have a very lean organic model that allowed us to survive to the point of commercialization. Now anyone with a very high burn rate is in a very different position, of course, but I think it’s important to not necessarily confuse market consolidation with technical and value consolidation–not anymore, as both are occurring, and can make sense in specific cases.

  4. Steve Ardire

    Hi Nova,

    Not sure about your article headline but for true realization of cognition as a service you need context driven dynamic algorithms for automating pattern discovery with common sense reasoning AND expressive intelligence that learns to infer, adapt, improve over time. Interesting you did not mention IBMWatson which invariably will hook up with Apple SIRI and new players like ( ex SIRI peps ) looking to create an intelligent conversational interface to anything

    PS – nice seeing you at Cognitive Computing Forum

  5. Kevin C. Brown

    Personas are important, but only if they are relegated to a minor role in the larger picture. Many more humans dislike anthropomorphic machines than care for them. Even the best anthropomorphic presentation gets old very quickly, making users feel as they are trying to communicate with an autistic savant.

    Much more important is the feedback loop from misinterpreted input or inadequate output. Most users just give up trying and the virtual assistant can’t learn from the exchange. This is where the jump from today to very useful virtual assistants will come from.

  6. Bob Sullebarger

    Great post, Nova. Ultimately, every B2C enterprise is going to need to field its own Virtual Assistant that works seamlessly across channels. From a technical perspective, containing the grammar and problem space addressed by any given IVA is definitely what makes it doable.

  7. Before consolidation though, we would need to build domain dependent intelligent virtual assistants for many different domains and make them robust. We can the use their cognition as a service to build individualized assistants. There are many similarities between the debate on future of education and this article – personalization is one of the primary focus today. While other domains are transactionary, education is more temporally contextualized and therefore need a different algorithmic structure. Similar peculiarities would exist in each domain.

  8. Matthias L. Jugel

    I dont’ think Siri counts as an assistant, just as many of the other speech-recognition based search engines. Whatever intelligence they have to figure out what you actually wanted to know, it’s still a voice commanded search engine.

    A real assistant will take actions on your behalf without command to assist you in doing what you may want to do. One that would come close was Donna, which was acquired by Yahoo! and basically killed.

    I have never since found another software like it.