The main challenge that organisations face in implementing machine learning is the complex infrastructure or workload needs. A whopping 90% of CXOs feel the same way. Into the details of this – 88% struggle with integration and compatibility of AI/ML technologies, while 86% struggle with the frequent updates that are required for data science tooling.
Such stats by the DataRobot 5 Latest Trends in Enterprise Machine Learning 2021 report state that many organisations do have a difficult time keeping up with ML. This implores the question – is ML really overhyped?
Every year, there are always some technologies that are more popular than others. This was seen with cloud computing, big data and cybersecurity. Machine learning is currently the topic that allows people to dream about the future and the possibilities that ML can introduce. The dreams are even so scary as they include self-learning robots that can take over the world. But the reality is so far away from this. Today, it is difficult to crack the working of statistical and mathematical supervised learning models that are deployed in machine learning.
Sign up for your weekly dose of what's up in emerging technology.
Such future dreams surely motivate us to invest in technology but also drive the so-called hype. Experts suggest that such situations arise when ML is requested without actually considering the internal data readiness or the requirements of the tool.
ML Engineers Struggle
A quick look at the answers of ML engineers and data scientists on Quora suggests that they are not satisfied with their jobs at places where there are ‘hyped’ expectations by companies. Some of the top arguments put across by the employees include:
- They were hired to do basic data analysis like the one that involves using excel sheets, R analytics, or analysis in Python. In any of these cases, ML did not exist at all.
- Data is sparse, and the company doesn’t have the right features to collect it. This causes a high degree of non-linearity, and the model comes back with a low accuracy rate.
- End up being an ‘SQL junkie.’
- Apprehensive managers who are not sure what ML can do, don’t fund experiments as they worry about businesses losing money.
- IT does not cooperate, sometimes even in sharing cloud passwords for complete access.
Becoming Data First
For deploying machine learning, it is essential to have a solid foundation of data for successful project execution. This demands a complete change in organisational processes and culture.
Enterprises must work on ‘data readiness’ before any machine learning development begins. This includes getting clean and stable data; and creating data governance processes as well as scalable data structures. The companies need to implement long-term data-based plans and policies to create a common data architecture.
Employees require time to adapt while onboarding any new technologies, and ML is no exception.
New technologies are always overhyped
When computers were gaining popularity in 1950, people thought that the future of these machines were humanoids – especially the military. But nobody imagined the Internet would actually change the world. Similar is the situation today, where the latest algorithms developed in AI and ML are always overhyped.
ML is not something that is very new, though. Arthur Lee Samuel, an American pioneer in the field of computer gaming and AI and who popularised the term “machine learning,” defined it as something that gives computers the ability to learn without being programmed explicitly in 1959. Today, ML is more objective in nature and is concerned with what can actually be achieved in realistic terms.
ML has been successful at doing many amazing things like production parameter adjustment, predictive maintenance, and visual quality control. It is essential to set an achievable long-term goal and work on organisational infrastructure, data strategies and culture. According to Paul Zhao, Principal Product Manager, Data Science and Machine Learning, Snowflake, “The need to leverage machine learning for better and faster insights is clear. Only organisations that are able to rein in the complexities around infrastructure, tooling, operations, and workloads will be able to deliver on the value of those insights.” The possibilities of what ML can achieve is endless, and it probably deserves the hype.