AI is real, AI is coming, but AI is complex

Reported by: |Updated: July 13, 2018

By 2020, 30% of the organization that fail to adopt AI will not be operationally and economically viable, said Pankaj Prasad, Principal Research Analyst at Garner India, speaking at Gartner IT Infrastructure, Operations Management & Data Center Summit 2018 in Mumbai. Every vertical will be impacted by AI in some form of the other.

One question everyone asks is whether AI is real or hype? AI is on the rising slope of expectations in the Gartner hypecycle. Machine Learning (ML) and Deep Learning (DL) are at the peak of inflated expectations. The world is at least 5-10 years away from commoditized applications of AI.

“When did you last think of AI in your pockets?” asked Pankaj alluding to use of AI technologies on smartphones. The scenario in business is different: “A survey among companies reveals that 14% show no interest, 35% have AI on their radar but have no action planned, 25% are planning for AI in medium or long term, 21% are planning in the short term or actively experimenting, and merely 4% have already invested in AI.”

Success Factors

Companies should focus on some common applications of AI, for example areas like root cause analysis, contextualization and co-relation to deploy AI successfully. The long-term goal is large scale AI enabled automation, but one should start small. Pankaj highlighted 5 key stumbling blocks: (i) Lack of skill sets (ii) Infrastructure scale (iii) Management complexity (iv) Increasing demand and (v) Tighter IT budgets.

91% companies say that data quality is main inhibitor of AI. Another important factor is that AI is heavily dependent on infrastructure. To calculate TCO accurately, one needs to account for the high cost of specialized hardware. He highlighted 6 factors that influence the choice of infrastructure: (i) Runtime flexibility (ii) Data gravity and integration (iii) Performance (iv) Operational simplicity, (v) TCO and (vi) Infrastructure choice.

Pankaj averred that one key question to ask in deploying AI should be “how to avoid failure”. To do that, IT must focus on low value repetitive tasks, ie, the low hanging fruit. This will free up human resource who can then upskill themselves to be able to frame problems and interpret the results. Bottomline is that IT teams will have to get into core AI and get their hands dirty as there is no out-of-box solution.

Vendors & Tools

Some of the emerging AI operations platforms (AIOps) Pankaj mentioned are Moog, ExtraHop, BigPanda and Graylog. They can integrate with other sources of data. They have visualization built in. They leave the IT team with 2 questions – ‘What are the data sources?’ and ‘What are the outcomes?’ Pankaj added that these tools need a lot of customization and data training in order to reduce both the false positives and the false negatives. For IT operations use case, AIOps is an area where there is a luxury of pre-built and ready to deploy platforms.

For the business use case, AI users will need a wide range of tools to handle the raw data. As an analogy, in the kitchen, different types of knives and food processing equipment are required.

Vendors tend to hide the key benefits of the AI layer. Pankaj proposed an important question that every organization must find an answer to: ‘How do you know whether you are getting actual AI or analytics disguised as AI?’

[email protected]