Awaken the Machines
Delwin Graham - Dec 16, 2017
Artificial intelligence and machine learning are new categories of computer science which transition computers from mere followers of instructions to machines capable of “thought”.
Artificial intelligence and machine learning are new categories of computer science which transition computers from mere followers of instructions to machines capable of “thought”. Of course, we have long considered and feared the possibility of machines thinking. In 1950, Allan Turing developed a test of a machine’s ability to exhibit intelligent behaviour equivalent or indistinguishable from that of a human. In the present case, machines are considered to be “intelligent” to the extent that they are able to improve performance, that is, to “learn”, without human intervention.
Machine Learning is arguably one of the most important General Purpose Technologies (GPTs) of our generation. (Cf. Daud Khan and Paul Morland, Software and Services: Industry Update; Canaccord Genuity Limited (UK) November 28, 2017). GPTs are technological innovations which have had a wide-ranging impact on multiple industries. Examples include the wheel, electricity, the steam engine, the Internet and the internal combustion engine. They all have reshaped the economy and ultimately boosted productivity across all sectors and industries. However, the ability of machines to continually improve with or without human intervention turns much of this technology landscape on its head. Technology has been driven by developers who codify actions which the machine then repeats faster and more accurately than a human could. However, machine learning tackles problems that are difficult to codify because we, as humans, struggle to understand how we get to a particular answer. For example, how do we recognize faces?
With the proliferation and retention of more data, machines can become more adept at recognising patterns in diverse sets of data to come up with better answers and ultimately a competitive advantage for users. Here are a few illustrations. In the case of cyber security, most anti-virus systems use signature -based techniques to block threats but these are becoming redundant with new attack sites (e.g., Internet of Things (IoT) devices, mobile, etc.) and methods using artificial intelligence. With its ability to learn quickly with a large set of ongoing test data, machine learning can make immediate decisions to enable a rapid response to prevent threats rather than simply reacting after the fact. One significant advantage over signature-based techniques is the ability to detect minor differences in the executed code. In the case of financial fraud detection, financial institutions typically rely on historic data to create a pattern of potential fraudulent transactions. Similiar to legacy anti-virus systems, this is signature -based protection that does little to detect first-time fraud. Also, the fraud models are often updated infrequently due to the cost and time required for accurate modelling, and this can lead to fraud schemes remaining undetected for long periods of time. The other concern for banks is to maximize customer satisfaction and thus minimise false positives, where legitimate transactions are flagged as fraudulent. By modelling and predicting individual behavior in real-time, machine learning can serve to detect a change in an individual’s behaviour and revaluate the risk, thus blocking new fraudulent attacks in real time and reducing false positives.
Machine learning is already being used to generate personalized customer offers. Amazon’s recommendation engine has huge success in driving incremental purchases by integrating it throughout the buying process, from product discovery, checkout and after sales. In fact, around 35% of all Amazon sales are estimated to be generated by the recommendation engine (source: Khan). Also, Netflix uses machine learning to access the content you watch and predict the content that you might enjoy. Google is using machine learning to improve its translation engine. One of the historical problems with machine translation is that it involves translating one word at a time and this can transform the meaning of the sentence. Google uses a machine learning system, i.e., Neural Machine Translation, which essentially tries to capture the higher meaning of a sentence by comparing each word with every other word so as to indicate whether there are any other meanings. In the case of online gaming/gambling, every click of a customer’s mouse, speed of reaction, and reaction to wins and losses can be plugged into a predictive analytics machine that in enhanced by machine learning, and can be used not only to enhance revenue but also to flag certain types of players like addicted gamblers.
Machine learning is happening now and will transform most industries. We can expect that there will be an acceleration of investment in this area in the coming years. Everyone will have machine learning strategies using supposedly high functioning algorithms, but the key to actionable results will be the quality and aggregation of the data. The key challenges will be the ability to aggregate diverse sets of data in relative near time and to apply the machine learning algorithms to detect patterns and suggest actionable ideas. Machine learning is at its best when fed on multiple sources of data where it can uncover patterns, see hidden information, and predict future decisions. The economics of bringing this data together suggests that it must be housed in the cloud, although the sources may be either cloud or non-cloud. The data sources are likely to be live (i.e., actively changing) and large scale. To this point, Oracle recently announced that it will launch the fist “fully autonomous” database cloud service. By incorporating machine learning, this new version will not require humans to update, manage or tune the database. This should materially improve the economics of running database systems, increase uptime and decrease downtime. As you can imagine, companies that control the data are looking to control the machine learning space. The FAANGs (Facebook, Apple, Amazon, Netflix, Google) and Microsoft have been active acquirers of smart algorithms.
Please contact me (firstname.lastname@example.org; (780) 408-1518) for further details and a few actionable investment ideas.
This article is solely the work of its author a registered Investment Advisor at Canaccord Genuity Wealth Management. The views (including recommendations) expressed in it are those of the author alone, and are not necessarily those of Canaccord Genuity Wealth Management. The information contained herein is drawn from sources believed to be reliable, but the accuracy and completeness of the information is not guaranteed, nor in providing it does the author or Canaccord Genuity Wealth Management assume any liability. Canaccord Genuity Wealth Management is a division of Canaccord Genuity Corp. Member – Canadian Investor Protection Fund.