Six reasons for the failure of first generation Enterprise AI
There is little doubt that Artificial Intelligence (AI) will transform industries and business models. Recent developments in driverless cars, voice recognition, and deep learning show how much machines can do and the powerful business models AI can drive.
Driven both by opportunity and a fear of losing out, companies across industries have announced AI-focused initiatives. However, early AI pilots are not producing the dramatic results that technology enthusiasts predicted and executives expected.
Here are six myths and design flaws driving failures in Enterprise AI:
1. AI is going to kill our jobs
The opportunity for augmenting jobs with AI is way larger than replacement of humans. AI, like every major technology innovation, will disrupt and dislocate some jobs no doubt.
However, while AI and related technologies are projected to kill 5 million jobs by 2020, there are 1.3 billion global workers whose jobs will be dramatically enhanced and improved through AI. Not to mention an entire class of "new collar" jobs that AI will bring about.
2. Big Data and Analytics are AI
Big Data and Analytics are cool things that computers do and are used in building Enterprise AI systems of intelligent engagement. Data and Analytics are critical for an AI but they are like your senses: “just because smells can trigger memories, it doesn’t make smelling itself intelligent, and more smelling is hardly the path to more intelligence.”
3. NLP, Machine Learning, and Deep Learning are AI
These are just tools and techniques for complex pattern recognition and similar to other tools such as algorithms for search engine and email spam filter. Comparing Enterprise AI systems to NLP or ML or Deep Learning is like comparing wrenches and fuel pumps to a car.
4. Robotic Process Automation or Chatbots is AI
RPA and chat bots mostly handle rule-based work and structured data inputs. They do not handle judgement-based work and unstructured data. They are themselves not intelligent and do not proactively provide personalized, contextual and explainable insights to humans as good AI systems should.
5. Enterprise AI can be an opaque Black box
Enterprise risk and compliance officers will not accept AI that is not explainable, makes important decisions without accountability, or worse is destructive. However, 99+% of AI startups operate AI as a Black box and have little to no hope of businesses taking their science projects into production (imagine defending a wrongful death claim due to a denied insurance benefit by just saying “my RNN/CNN/DNN told me so and I don’t know why!).
Therefore, enterprise adoption of AI will require a new class of capabilities that are not being talked about enough today. These include AI that is explainable, AI that builds trust by providing evidence, AI that can help you manage bias, safe interruptability, and a handful of other things.
6. Technology alone will solve all problems and overnight
Enterprise AI is about building, deploying and managing systems of intelligent engagement. It is not about mere automation of manual work. It requires a new class of technology, data, methods, and skills focused around a business led portfolio of industry optimized AIs and not just point solutions or science projects.
The Road Ahead:
Going forward, businesses that focus on leveraging AI as a strategic business capability to augment and amplify human and process intelligence stand to gain the most. Doing so requires a careful balancing of technology, data, processes, and skills all looked through and implemented through an industry and process lens.
About the Author:
He is also a founding managing director of The Entrepreneurs’ Fund IV, a $100m seed fund focused exclusively on the cognitive computing and machine intelligence market with eight active investments. Portfolio companies includeSpark Cognition, a cognitive security analytics company where Saxena is Chairman of the Board and WayBlazer, a B2B AI platform for the travel industry where Saxena is a Board member.
Previously, he was IBM’s first general manager of IBM Watson (2011-14), where his team built the world’s first cognitive systems for healthcare, retail, and financial services.
In his spare time, he competes in long distance automobile races in pre-war, classic, and contemporary cars. More at www.saxenaracing.com on his upcoming 28 day endurance race in October and November across southern Africa in a race modified 1934 Alvis Speed 20.