Z Digital Agency Manager interviewed Aaron Richiger, Co-Founder at Turicode, to enquire about the current state of the Artificial Intelligence Industry, beyond the buzzyness. Turicode is digitalizing your company’s contents to automatically create websites, mobile apps and knowledge bases, with machine learning.
If you had to summarize the current state of the AI industry in 2 minutes, what would you say?
Let’s start with the bad news: AI and machine learning are hyped buzzwords, or as Robin Hanson said: “Most firms that thinks they want advanced AI/ML really just need linear regression on cleaned-up data”. People without experience in AI/ML are very likely to expect too much of an intelligent system, e.g. from a chatbot or a deep-learning based language translation engine. Or as a friend of mine uses to say: “We spend half of the time in our AI projects with customer-expectation-management”…
Having said that, good news remain: Recent advances in algorithms, hardware and easy-to-use ML frameworks enabled solutions for tasks that seemed impossible for an artificial intelligence a few years ago. And I predict that we have barely scratched on the surface of the full potential of AI.
Therefore I am optimistic, that novel state of the art solutions will meet the high customer expectations in the near future.
What advantages do you see in the AI methods you’re currently working with compared with other methods?
AI solutions have the following advantages over traditional rule-based solutions:
- Scalability: AI systems scale/generalize better for never-seen/unexpected inputs.
- Simplicity: For complex tasks, the AI approach might be simpler, due to the huge number of necessary rules for a rule-based approach. Hand-crafting rules is time-consuming, costly, error-prone and scales badly in the case of unexpected variances in the future input data.
- Complexity: AI (i.e. deep learning) is able to solve complex tasks that are impossible to solve with a traditional rule-based approach (e.g. natural language translation).
What are the limitations you foresee in the development of AI?
We experience the two following major difficulties in our daily work with ML algorithms:
- A huge amount of training data is necessary for training an AI, but annotated training data is often not available and costly to generate.
- An AI has similar limitations to human intelligence, e.g. if humans don’t agree anymore whether the animal on a picture is a cat or a dog due to ambiguity, the AI will have similar difficulties to make a correct decision.
Any recommendation about a person, community or a great information source about AI?
Instead of recommending a person or community, I would like to highlight the achievements that easy-to-use ML frameworks such as Keras and Theano have introduced.
What is currently your most important personal challenge?
Finding customers for our technology: We have working solutions, now the world needs to know about them…
