Siri, Watson, AlphaGo, Cortana, Alexa and Deep Blue, what do
these names have in common? Yes, they
are notable Artificial Intelligence (AI) projects or products. AI is becoming more and more important. From chess playing to personalized
advertising; voice recognition to stock market prediction, AI seems to be
capable of taking over human’s day-to-day activities. Is
this true?
The following lists show the tasks that AI can and cannot do
(yet):
AI can:
- Play Chess (obviously!)
- Face and speech recognition (go try your Facebook tag feature)
- Personalised shopping recommendation (you plan to buy one item from Amazon but may end up buying more from the “people who bought this also bought that…” recommendation)
AI cannot (yet):
- Play football (imagine someday AI robots winning the World Cup!)
- Automated scientific discovery (AI thrives on existing theory, technology and data)
- Create new product for shoppers (AI could spot the prominent features of popular products but may not create a brand-new product from the data)
A report by Mckinsey&Company highlighted 5 limitations
of AI as follows:
1.
Data labelling
·
AI is “trained” by feeding tonnes of data into
its machine learning capabilities. Humans
need to label the data correctly in order for the AI to interpret. Bad or wrongly labelled data will lead to
unreliable outcome.
2.
Obtaining massive training data sets
·
AI’s reasoning capabilities are based on the
results of analysing past data. Larger
data sets would improve the accuracy of AI’s response.
3.
The “explainability” problem
·
The results concluded by AI may be hard to
explain in human terms, especially when it was reached in real time. Shall users just take the decision as it is?
4.
Generalizability of learning
·
Unlike the way humans learn, AI models have
difficulty carrying their experiences from one set of circumstances to another.
Companies must repeatedly commit
resources to train yet another model, even when the use cases are very similar.
5.
Bias in data and algorithms
·
Human predilections (conscious or unaware) are
brought to bear in choosing which data points to use and which to disregard.
Furthermore, when the process and frequency of data collection itself are
uneven across groups and observed behaviours, it is easy for problems to arise
in how algorithms analyse that data, learn, and make predictions.
As AI is an on-going development. The above limitations could be resolved soon. We found a video from internet about human vs
robots in Football match, enjoy!
No comments:
Post a Comment