The argument that eventually machines will render human input in the machine learning process redundant is inaccurate. Machines require vast amounts of training data to achieve competency in tasks that humans can process quickly and easily.
Appen CEO Mark Brayan and Figure 8’s VP of product management and design, Alyssa Simpson Rochwerger, hosted us at Figure 8’s San Francisco office. Figure 8 is a typically San Franciscan tech business, filled with bean bags, musical instruments, greenery, gaming consoles, casually dressed staff and even a couple of friendly canines roaming the floor. Despite the casual air, there was also a clear sense of enthusiasm, entrepreneurial spirit and an understanding of the significant role Figure 8 plays in the rapidly evolving field of machine learning and the artificial intelligence applications that derive from it.
One of the key messages from our time in the Figure 8 offices was that the quality of a machine learning algorithm is directly correlated with the quantity and quality of the training data used to develop it.
Machines need human intervention to learn
Machines require vast amounts of training data to achieve competency in tasks that humans can process quickly and easily. Examples include using machines to perform visual inspection tasks such as scanning supermarket shelves for diminished inventory or spotting paint blemishes on an automotive production line. For a machine to perform visual inspection tasks, such as these with low error rates, the algorithms that drive them require exposure to the many situations they are likely to encounter, a lot of which will be subtle and nuanced – the so called ‘edge cases’ – that require high-quality human annotated data.
The opportunity set of real world problems with narrow and specific use cases requiring visual decisions is only limited by our imagination. To solve these problems at massive scale, whilst ensuring high quality and low cost, requires a combination of machine annotation and human input for the ‘edge cases’ that an algorithm has not yet been trained to identify. In combination, Appen and Figure 8 are uniquely positioned to solve many of these evolving machine learning challenges by drawing on the automation and speed of machines as well as the power of the human mind to quickly identify and tag the more nuanced cases.
The argument that eventually machines will render human input in the machine learning process redundant is inaccurate. Machines can deliver high levels of confidence in the predicted outcome for a given set of inputs in a majority of cases but to achieve the highest levels of confidence in all possible scenarios – returning again to the concept of the ‘edge case’ – requires huge amounts of high quality training data that a machine cannot generate on its own. Is 98% confidence in 98% of likely scenarios good enough? It is until the machine fails when faced with an edge case. Depending on the application, failure can result in catastrophic outcomes both in economic terms and with respect to human life (autonomous vehicles, for example). The potential for catastrophic failure sits at the core of underlying trust and transparency concerns with AI – and the only way to solve for it is through the application of large volumes of high quality training data.
Leadership in machine learning
Figure 8 is helping democratise the development of applications based on machine learning through its unique and scalable customer platform. Customers can self-serve, keeping costs low, as well as access a library of templates for sourcing training data. For applications that are amenable to automation – typically visual and some audio – machines can accelerate the data annotation process and for situations for which automated annotation is less applicable – content relevance, for example – customers can access the services of the human ‘crowd’. For smaller organisations, Figure 8 provides access to large-scale machine learning infrastructure at relatively low cost.
Appen is currently focused on three strategic investment areas:
- Annotation efficiency – Achieving the lowest possible unit cost whilst maintaining high levels of quality is a key focus for the business.
- Flexibility – Annotate more types and more complex data. Fostering a culture of agility that allows the company to adapt to new data types as they emerge.
- Quality – continual improvement.
Consistent with Appen’s business flexibility focus is its evolving capability in the supply of data annotation services in highly secure environments, particularly for government clients. Appen currently runs two secure facilities and this will probably grow over time. Appen is also working on the development of a more secure ‘at home’ data annotation service.
A truly global addressable market
Appen is looking at broadening its business focus beyond the US with an initial focus on growing its existing operations in parts of Europe and Asia. Figure 8’s platform helps in these regions as it offers a lower price point. Longer term, China is the big prize however navigating the operational, cultural and political challenges China presents will be a significant task. Protecting the company’s intellectual property is paramount. In a country that has publicly stated its desire to be a global leader in the field of artificial intelligence the risk of IP theft or replication is presumably high. To this end, the Chinese operations will have their own, separate, technology stack. This will allow the Chinese team to go as hard as they can at developing their own capabilities without putting Appen’s core technology at risk.
The current anti-trust issues swirling around some of Appen’s largest customers in the US probably presents an opportunity for Appen as these customers seek to avoid the worst case outcome of corporate break-up. One of the issues the large US tech companies need to manage is how they use customer data. This is an area where Appen believes its services can be helpful, particularly their secure offerings, with respect to improving biased data models and in the development of better algorithms that deliver superior content delivery outcomes or through the proactive identification, management and removal of malicious content.
Appen brings a far more rigorous approach to the development of the economic model within Figure 8. Appen brings an ‘obsession’ with unit economics to the broader group and should help Figure 8 successfully transition from start-up to a strong and profitable business. At the same time, Appen is conscious of fostering the entrepreneurial culture within Figure 8 and will also seek to leverage the inherent advantages this brings to the combined group. An example is driving a shift in the sales culture from ‘hunters’ to ‘farmers’ – that is, recognising and targeting the high incremental value derived from securing additional work from existing customers rather than the prevailing focus on securing new customers with all the associated customer acquisition costs and higher risk of churn. Other examples include avoiding very small customers for which the lifetime value to cost of acquisition ratio is unfavourable and encouraging the monetisation of the company’s professional services capabilities.
Access the innovative world of Emerging Companies
Disclaimer
The information contained in the article is given by Ausbil Investment Management Limited (ABN 2676316473) (AFSL 229722) (Ausbil) and has been prepared for informational and discussion purposes only and does not constitute an offer to sell or solicitation of an offer to purchase any security or financial product or service. Any such offer or solicitation shall be made only pursuant to a Product Disclosure Statement or other offer document (collectively Offer Document) relating to an Ausbil financial product or service. A copy of the relevant Offer Document may be obtained by calling Ausbil on +612 9259 0200 or by visiting (VIEW LINK). You should consider the Offer Documents in deciding whether to acquire, or continue to hold, any financial product. Neither this article nor the provision of any Ausbil Offer Document is, and must not be regarded as personal advice or a recommendation or opinion in regards to an Ausbil financial product or service or securities of any other entity including Figure 8 or Appen or that an investment in an Ausbil financial product or securities of any other entity including Figure 8 or Appen is suitable for you or any other person. This article and the information it contains is for general use only and does not take into account your personal investment objectives, financial situation and particular needs. Ausbil strongly recommends that you consider the appropriateness of the information and obtain independent financial, legal and taxation advice before deciding whether to invest in an Ausbil financial product or service or in the securities of any other entity including Figure 8 or Appen .The information provided by Ausbil has been done so in good faith and has been derived from sources believed to be accurate at the time of completion. While every care has been taken in preparing this information. Ausbil make no representation or warranty as to the accuracy or completeness of the information provided in this article, except as required by law, or takes any responsibility for any loss or damage suffered as a result or any omission, inadequacy or inaccuracy. Changes in circumstances after the date of publication may impact on the accuracy of the information. Ausbil accepts no responsibility for investment decisions or any other actions taken by any person on the basis of the information included. Past performance is not a reliable indicator of future performance. Ausbil does not guarantee the performance of any Fund or the securities of any other entity including Figure8 or Appen, the repayment of capital or any particular rate of return. The performance of any Fund depends on the performance of its underlying investments which can fall as well as rise and can result in both capital gains and losses. By viewing this article, you agree to be bound by these limitations, terms and conditions set out in the paragraphs above. Unless otherwise specified, any information contained in this publication is current as at the date of this article and is prepared by Ausbil Investment Management Limited (ABN 26 076 316 473 AFSL 229722) (Ausbil). Ausbil is the issuer of the Ausbil MicroCap Fund (ARSN 130 664 872) (Fund). This article contains general information only and the information provided is factual only and does not constitute financial product advice. It does not take account of your individual objectives, financial situation or needs. Before acting on it, you should seek independent financial and tax advice about its appropriateness to your objectives, financial situation and needs. Securities and sectors mentioned in this monthly report are presented to illustrate companies and sectors in which the Fund has invested and should not be considered a recommendation to purchase, sell or hold any particular security. Holdings are subject to change daily. The value of an investment and the income from it can fall as well as rise and you may not get back the amount originally invested. Past performance is not a reliable indicator of future performance. Unless otherwise stated, performance figures are calculated net of fees and assume distributions are reinvested. Due to rounding the figures in the holdings, breakdowns may not add up to 100%. No guarantee or warranty is made as to the accuracy, adequacy or reliability of any statements, estimates, opinions or other information contained herein (any of which may change without notice) and should not be relied upon as a representation express or implied as to any future or current matter. You should consider the Product Disclosure Statement which is available at (VIEW LINK) before acquiring or investing in the fund.