The Evolution of Machine Learning in Business​

 

To gain a competitive edge today, companies have infused Machine Learning (ML) to automate ever-increasing streams of data, enable data-driven decision making and drive real-time business value. Though ML was first introduced in 1950s, you may be surprised to know ML only took off in the business world in the last decade.  If you want to better understand ML and how your businesses can capitalise on ML, look no further! In this article we will present an overview of the past, present, and future of machine learning in businesses:

1. What is ML?
2. Before ML coming into picture: statistical modelling
3. The evolution of ML
4. What is the current state of ML in business?
5. What’s the future of ML in business?

1. What is ML?

Before we dive into the history of ML, let’s talk about what this buzzword is. ML is an application of Artificial Intelligence (AI) that allows machines to automatically learn and improve from experience without being explicitly programmed. ML models use statistics to find patterns in large amounts of data (e.g. numbers, words, images, clicks) to make predictions. These ML models then learn and adjust continuously based on how accurate the initial predictions are. Figure 1 below illustrates general ML development cycle: 

Figure 1: Process of building ML models (Source: yourtemplate.com) ML-enabled predictions can already be found in many areas of business operations. From bank fraud detection to customer personalised recommendation (this is how Netflix can always recommend movies you like), organisations use ML to provide better customer experience, improve operational efficiency and increase revenue streams.

2. Before ML Coming into Picture: Statistical Modelling

ML would not exist without statistics. Before ML coming into the picture of predictive analytics, organisations used statistical modelling to transform data into business insights.  Statistical modelling is a method of mathematically approximating the world. Statistical models are designed to find relationships between variables and the significance of those relationships, whilst predicting future values. For example, marketers use statistical modelling to divide customers into different segments based on various factors (e.g. priorities, demographic information, needs), so that they can implement specific marketing strategies for different segments. 

Though being handcrafted with high level of interpretability, statistical models have limited predictive accuracy, since sometimes the underlying assumptions of the model are far too strict to represent reality. Today’s businesses are adopting hybrid methods combining characteristics of statistical modelling and machine learning, so as to understand in-depth how the underlying models work as well as generate accurate predictions. 

3. The Evolution of ML  

Now, let’s take a short walk through the history of ML.

1950-1980 ML in the lab

These years are the very early stages of ML. In 1950, Alan Turing created Turing Test to determine whether a computer can think like a human being. Two years later Arthur Samuel of IBM first came up with the phrase “Machine Learning”. In the following decades, scientists created the very first ML programs for simple applications, such as improving computer’s performance in a game checker and recognising rough pattern and shape. 

1980-2000 From Lab to Reality

By 1982, interest in neural networks (a subfield of ML that deals with algorithms inspired by the structure and function of the human brain) started to pick up again. In the last several years of 1990s, many more advances have been made thanks to the increasing availability of digital data from internet growth. ML industry shifted from a knowledge-driven approach to a data-driven approach. In 1997 IBM’s Deep Blue, a chess-playing computer, beats the world champion at chess. Since then ML leaves the lab and gets down to reality.

2000-2010TriggersBusiness Interest 

The beginning of 21st century saw more and more businesses shifting their attention towards ML. Tech companies like Google started realising ML’s potential in applying complex mathematical calculations to big data. They are investing considerable resources and researching heavily in the field to stay ahead of their competitors. In 2014, Google acquired two-year-old AI startup Deep Mind for $500M, making it Google’s largest European acquisition to date. 

2010-2020Modern Machine Learning 

In the past six decades, ML was poised to take off in business world because of limited datasets and expensive computing power. These barriers burned away in 2010s and ML finally ignited powerful changes in real world. Symbiosis of data collection and cheaper memory allows storage of massive amounts of data. Faster processing power enabled by GPU (Graphics Processing Unit) manufacturers like Nivida makes applying machine learning practical. In addition, tech giants have fueled the expansion of ML into various sectors, such as Google speech recognition and Facebook DeepFace facial recognition.

4. The current state of ML in business  

According to Algorithmis’ survey “2020 state of enterprise machine learning”, the top three objectives for which companies of all sizes are adopting ML are: 

• Reducing company costs
• Generating customer insights and intelligence
• Improving customer experience

Currently, the most common ML applications are customer-centric, spanning everything from improving in-store retail experiences with IoT to boosting security with biometric data to predicting and diagnosing disease. The table below presents some examples of ML use cases in different industries that are driving business value today. 

“The future is already here. It’s just not evenly distributed.”—William Gibson    Although ML is being implemented more widely due to its proven value, it is not equally adopted across businesses. Only big companies which can hire scarce data science and ML talents, and invest enormously in sophisticated IT infrastructure, are greatly benefiting from ML. In addition, ML adoption in business faces other challenges such as data management, model explainability, use cases identification, internal change resistance and ethical concern. 

5. What’s the future of ML in business 

ML is a continuously developing practice. In the future, ML industry will be independent from sophisticated technical framework and expertise, enabling ubiquitous adoption across different industries and companies of all sizes. Meanwhile, innovative ML methods will empower advanced analytics as well as increase energy efficiency of ML development and deployment. 

ML development tools: Automated ML for Everyone 

Though ML has enormous potential in business application, most companies are still in the nascent stages of ML adoption. Like Business Intelligence and other tech industries, we can only see exponential growth of ML business adoption when core ML development platforms and tools become affordable and available to every company. 

Tech giants and leading AI tool startups are automating ML to democratise ML for everyone. Microsoft Azure, Amazon SageMaker, Google AutoML are cloud-based ML platforms that enable data science professionals to build models and operationalise ML insights. Google introduced AutoML-Zero in a paper early this year, which shows that automatically building ML algorithms from scratch is possible by using evolutionary techniques.  

TurinTech has progressed further in this evolutionary field with over five years’ research. Inspired by Darwinism, the biological evolution theory, TurinTech’s Evolutionary AutoML platform helps users with different backgrounds access advanced ML to automatically create world-expert solutions.  However, there are some critical problems to be solved before Automated ML can really dominate the future: 

• Deployment: ML applications need faster lifecycles since data changes so rapidly in real business world that companies might need to deploy a new model every day. And businesses need ML development tools that can easily integrate with their existing systems.
• Data Privacy and Security: In many business applications, effective ML models are usually trained on sensitive data protected by strict regulations. Privacy-preserving ML is ever needed to ensure data security and privacy.
• Explainability: ML models are often seen as black box for their lack of transparency. In businesses, knowing the “why” is as important as predicting the “what”. Unless ‘why’ an ML model made a decision that can be explained to stakeholders, its applications will be strictly limited despite achieving higher accuracy than traditional methods.
• Model validation: Data is changing fast in business environment. ML models need to be validated that they can predict accurately on new data.
• ML ethics and bias: ML and the training data are necessarily biased.  Knowing the different bias present in ML life cycle and the potential consequences requires a transparent process. Reducing or even avoiding discriminatory bias require bias testing in development cycle as well as monitoring and reviewing models in operation.

At TurinTech, our evolutionary approach enables faster ML creation and deployment, using a transparent end-to-end process (white-box), while keeping business confidential without modifying users’ underlying data and model. 

ML business use case: advancing to prescriptive analytics 

Advanced ML technologies will not only predict what will happen in the future and explain the underlying reasons, but also recommend the optimal solution for the best outcome. Advanced ML evaluates each available option in complex business environments with millions of variables and constraints to achieve customised business objectives such as profits and operational efficiency .  TurinTech uses evolutionary optimisation techniques to empower businesses to make optimised value-driven decisions in a cost-effective and effortless approach. Organisations can easily solve complex business problems while coping with millions of trade-offs.  

Green ML: ML with low environmental impact 

A university research in 2019 found that training an ML model can emit more than 600,000 pounds of CO2. That’s almost five times the amount of CO2 emitted by the average car during its lifetime. Efficient ML methods are demanded to build a greener future.  MIT researchers have been developing new ML techniques that adapts algorithm to various hardware to reduce the computing power required during training. TurinTech’s solution to this problem is to automatically optimise AI models and software code to achieve optimal energy efficiency both during model training and in production (we will further discuss this topic in our future blogs). 

Conclusion 

ML has come a long way since 1950s and finally risen to prominence in recent years. This has been made possible with easier availability of Big Data, cheaper storage technologies and research advancements led by tech giants.   ML is transforming businesses today across different sectors. Anything less than a data and ML-driven culture is ripe for failure. However, the future of ML is not evenly distributed. In addition, more advanced ML that recommends optimal solutions and greener ML techniques are called for in the future business world.  TurinTech aspires to empower every business to have equal chances in accessing and capturing value from advanced ML. We have built a next-generation AI optimisation platform to automate the entire lifecycle of creating and deploying AI solutions. Our evolutionary approach enables businesses to make optimised decisions in complex environments and build energy-efficient ML solutions.  Learn more about how your enterprise can optimise with advanced AI at https://turintech.ai. Follow us on LinkedIn, Medium, Twitter.

References:

1. https://algorithmia.com/state-of-ml
2. https://huyenchip.com/2020/06/22/mlops.html
3. https://www.technologyreview.com/2018/11/17/103781/what-is-machine-learning-we-drew-you-another-flowchart/
4. http://ivyproschool.com/blog/2015/08/18/common-statistical-models-used-in-predictive-analytics/
5. https://towardsdatascience.com/the-actual-difference-between-statistics-and-machine-learning-64b49f07ea3
6. https://www.kdnuggets.com/2019/08/statistical-modelling-vs-machine-learning.html
7. https://www.forbes.com/sites/bernardmarr/2016/02/19/a-short-history-of-machine-learning-every-manager-should-read/#121e9b7f15e7
8. https://www.dataversity.net/a-brief-history-of-machine-learning/
9. https://www.doc.ic.ac.uk/~jce317/history-machine-learning.html
10. https://www.verdict.co.uk/what-is-machine-learning-in-business/
11. https://mkomo.com/cost-per-gigabyte-update
12. https://www.forbes.com/sites/sap/2020/03/09/the-power-of-artificial-intelligence-vs-the-power-of-human-intelligence/#56ffb529346c
13. https://www.forbes.com/sites/davidteich/2018/12/26/machine-learning-and-artificial-intelligence-in-business-year-in-review-2018/#153ba8042041
14. https://www2.deloitte.com/us/en/insights/focus/cognitive-technologies/state-of-ai-and-intelligent-automation-in-business-survey.html
15. https://www.technologyreview.com/2018/11/17/103781/what-is-machine-learning-we-drew-you-another-flowchart/
16. https://ai.googleblog.com/2020/07/automl-zero-evolving-code-that-learns.html
17. https://www.7wdata.be/data-analysis/machine-learning-becomes-mainstream-how-to-increase-your-competitive-advantage-by-nidhi-chappell-intel-and-ronald-van-loon-adversitement/
18. https://www.geberconsulting.com/blog/machinelearning
19. https://towardsdatascience.com/the-state-of-ai-in-2020-1f95df336eb0
20. Onezero medium
21. https://www.forbes.com/sites/forbestechcouncil/2020/07/02/where-insights-meet-privacy-privacy-preserving-machine-learning/#2e41b8ad7332
22. https://hub.packtpub.com/machine-learning-ethics-what-you-need-to-know-and-what-you-can-do/
23. https://towardsdatascience.com/understanding-and-reducing-bias-in-machine-learning-6565e23900ac