EIGHT REASONS WHY AI MAY KILL SMALL BUSINESS*

© 2021 by Philip Borden

Claims for the power of artificial intelligence (AI) multiply every day. Small business owners find themselves challenged to keep up. Should they be trying so hard? Are there opportunity costs for doing so? Is the application of AI for small business a potential blessing or curse?

WHAT DOES AI MEAN TODAY, AND SO WHAT?

Artificial intelligence (AI) programming began in the late 1950s as an examination of the brain as a computer made of meat. Modeling the brain meant creating programs that imitated neurons. Computer scientists also applied their models to controlling robots. Given the state of computing power, neurons proved too complicate to model and by the 1990s, without supervision no robot could unpack the box it came in.

Within a decade of the invention of the term AI more practical solutions focused not on how brains worked, but rather how experts solved problems. Developers called this form of AI “expert systems.” Instead of lockstep algorithms that had to be complete and perfect, expert systems produced imperfect results quickly. They allowed users to progressively home in on intelligent solutions. Expert systems were simpler to implement, required less computer resources, and for much of the last thirty years of the twentieth century dominated AI thinking and practice.

Diagnostic systems from medicine to mechanics adopted them. They made inroads in inspection, quality control, troubleshooting, and training regimes. At first expert systems could be used only by large businesses that owned significant computing power. But advancing hardware technologies came to the rescue of AI and small business. Expert systems achieved celebrated results, but could only within a narrow band of tightly defined applications where expertise made a significant difference. They exhibited little learning and not much “common sense,” both earmarks of general human intelligence. By the late 1990s, the term artificial intelligence had become merely a programming style, and in some quarters, a term of derision. Artificial intelligence needed to return to its roots if it were to be considered intelligent at all.

Anew generation of AI developers took advantage of the fast- growing power and efficiency of computing by returning to the original idea of how the brain worked. They called their approach to building neurons rather than experts “neural networking” when describing how the software was structured, and “machine learning” when describing AI’s method of acquiring knowledge. Working with mechanical engineers, software engineers applied machine learning to the underlying structure and not just the controlling systems of robots.

Even when human teachers direct the neural training process, once set in motion today’s AI programs no longer require humans to guide them. They can self-assemble into new networks to accomplish complex tasks that require coordination beyond the abilities of individual robots. Nor are robots now limited to running mechanical devices. “Bots” can search through the Internet to find things it has learned to look for.

New AI tools have automated and deepened the learning process, sharpened the accuracy of “synapses” and accelerated the time to achieve results. They have gone beyond imitating voices to understand speech. They can direct large machine processes and modify them in real time. AI is replacing such soft skills as newspaper reporting and reacting — even emotionally — in communicating with and managing kids. AI is moving toward the general intelligence it originators once hoped for. Whether it will succeed is a much-debated question.

INFORMATION: AI’S MISSSING PIECE

These changes have profoundly amplified government and commercial power. Still, one more important barrier had to be jumped in order to boost AI onto its current higher plane: better data to learn from. The digital and social media revolutions created the ability to collect kinds of data unimaginable just a couple of decades earlier: torrential streams of sound and pictures and numbers and print and even gestures and direct brain wave measurements from personal and commercial websites, government collections, machines and industrial processes, television, DNA, and much much more. It comes in verified and unverified, simple and complex chunks. By 2005, the term Big Data (BD) had been coined to describe the amount of data available for computers to analyze.

We now measure data in zettabytes, chunks of one sextillion bytes. It is the amount of information space need to store the music for about 200,000,000 years of listening time. Every two days we create as much data as we produced from the beginning of time until the year 2000. Last year we created 2.3 zettabytes of data. We now have access to 50 zettabytes annually and soon will be in the yottabyte (1,000 times larger than a zettabyte) era. BD is itself only a step along the digital path to Really Big Data used in scientific studies about the genome and the biosphere.

BD has linked with “Big Algorithms”, formulas for analyzing and making use of data. Big Algorithms (BA) repeat, learn from the data, tweak themselves on the fly to adjust to new insights, and repeat again. BA has the same learning characteristics as AI’s neural networks or machine learning. BA is simply AI writ large. Today, data is often the most important thing determining a company’s worth on a balance sheet. Managing it a company’s largest challenge.

BD/AI is stunningly good at pattern recognition. It makes it possible to identify relationships that were previously unknown or hidden. However, neural networks operating on biased data becomes fact and AI still cannot decipher the meaning of the patterns it recognizes. For small businesses, the major difference between expert systems and neural networks is that the amount of resource the latter requires has placed it beyond the reach of most small businesses.

AI AND SMALL BUSINESS: THE GOOD NEWS

From the outset the military saw the potential benefit of AI. It pumped significant dollars into research but did not know how to make AI appropriate to the real time battlefield, training, and analytic problems it faced. So the federal government emphasized the importance of “dual use technologies” that originated in military research and could be commercialized, and vice versa. It sought companies large and small to help. In 1982 it launched the Small Business Innovation Research (SBIR) program, requiring federal agencies already paying for research by large companies to make 1.5% of their budgets available for innovative small business research. It offered financial support for the early development of new ideas, expecting that as small businesses developed successful products they would attract private funding.

A significant portion of the early technologies funded by the SBIR program involved expert systems. The military (and other government institutions) used expert systems to enhance training, pass routine decision-making down to lower ranked personnel, adjust battlefield strategies, and make quicker and cheaper equipment repairs. These uses were relatively easy to transition to commercial practice. Aided by the growth in hardware speed and capacity, the business community realized economic benefits. The term AI became part of the business vocabulary and important part of commercial public relations and branding.

SBIR funding has grown to 3.5% of contracted government research. Thanks to the SBIR and related programs which piggybacked on its successes, over its first fifteen years SBIR winners contributed more to advance technology research than all large companies and university research programs combined, an astounding but verified statistic. 70% of SBIR winners interviewed in 2018 indicated they would not have existed without the program and/or that the program significantly benefited their growth. In the period of 2014–2018 SBIR winners contributed over $92 billion in revenues and nearly $9 billion in federal taxes. They also created over 30,000 direct jobs annually, and tens of thousands more indirectly.

Despite considerable literature on the background and early history and analysis of the results of SBIR, there is little on the more recent success of AI development. None of the major studies of SBIR effectiveness single out software, much less AI. However, a search of recent SBIR grant winners shows no “new AI” studies, though several indicate expert system type support systems for management and research, and an increasing number focus on Bayesian analysis (a sort of fluid probability technology open to learning and change in a near AI way).

Although the SBIR program has garnered more respect, SBIR’s achievements have come at a price for small business. SBIR has had to share the stage with the Small Business Technology Transfer (STTR) program. STTR offers universities similar benefits, despite the fact that for many years private small companies, not universities, produced the most technological growth. Also, funding mechanisms have changed to allow large companies and financing concerns to participate earlier in the process, with a view to quick acquisition of small business technologies. This is especially true in the pharmaceutical space. Small companies no longer form the vanguard of AI in the SBIR program, certainly not the way they did in the expert system era.

While the SBIR program was transformative and helped create the fastest growing companies in the small business sector, it was only part of the small business story. Most small businesses remain life-style enterprises whose use of AI technologies continues to be marginal. Setting aside our Covid Year, since the opening of the AI era, small businesses have forming and failing at about the same rate. 81% of them have no employees, a slight uptick over the past three decades. It suggests that the universe of those potentially poised for rapid growth are shrinking even as software technology gathers steam.

For the 20% or so of small companies with any capacity to utilize AI growth strategies no matter how limited, here are the ways that it can do so, at least in theory:

§ New product development: aid in the technical research and research evaluation needed to create new products and services, and in the manufacturing processes to realize them through machine learned and expert analysis and control applications;

§ Marketing and promotion: identify customer needs/wants/predilections/preferences, and tailor advertising to specific individuals in the process, using conversational AI and BD/AI analysis;

§ Personnel: improve hiring practices by automatically evaluating large databases of resumes to match applicants with particular company priorities by reading and comparing applicant data and using BD/AI to evaluate their public histories;

§ Operations: support back office and logistic functions using the full range of BD/AI tools;

§ Customer service: identify product and consumer service issues, even probe customer complaints and concerns to go beyond simple reporting, drawing on both pure BD plus language recognition, translation, and speech generation;

§ Online presence: for retail and other purposes as appropriate, aiding in sales processes and image maintenance, using all the BD/AI tools.

Many of the functions that claim to be AI based do not use the “real thing” but only the name as a branding tool. For example, retail response to customers can be based on traditional decision tree programming, not learned or “intelligent” responses. Matching resumes to need does not have to be an AI function, and most practices are simple matching programs. In the end, the primary benefits of these traditional and sometimes actual uses of expert systems and BD/AI can create 24/7 service without the need for multiple full personal shifts.

AI AND SMALL BUSINESS: THE NOT-SO-GOOD NEWS

The benefits of AI come with costs that can be measured in investment dollars and organizational dysfunction. One expert has written, “Despite the seeming proliferation of artificial intelligence technology today, it is still in the very early stages of development. Contrary to what might be hyped on social media, in nearly all application areas AI is an expensive and complex solution without evidence of direct return on investment (ROI). This is especially relevant for small businesses with limited data, limited resources, and limited data science talent.”

§ AI can be costly. In the era of BD, collecting the requisite data, evaluating its integrity, storing it, and maintaining it in a secure environment has both initial expense and ongoing maintenance costs that many small businesses cannot afford. And BD/AI systems generate fundamental programming development costs as well as the ongoing expense attendant on a system that needs to keep learning to exist. In the post Covid-19 world, the loss of working capital reserves amplify the financial dimensions of the cost of AI for small business.

§ Not all AI solutions pay off in the short run. Hence, it can have a negative impact on both profitability and valuation. Moreover, focus on the development of AI solutions in non-software companies exacts both personnel and potential opportunity costs.

§ Poorly executed to ill-defined AI solutions can create reputational costs that translate into dollar and cent liabilities that can be hard to overcome. They also can generate a backlash among consumers who need “real human” connections for flexibility and response.

§ The ROI of AI solutions for small business has not yet been determined. Given the costs cited above it is not likely to be good, especially in the short run. And for an increasing number of small businesses, the short run may be its only run.

§ There is a natural cultural mismatch between the personal service and hands-on orientation of small business and efficiency drive of large business. Hence, the change to implementing AI solutions may be seriously negative. Even where the economics of AI may pay off, how it affects the staff, internal arrangements, and external image of the small business can hobble or even negate any realized gains.

AI AND SMALL BUSINESS: ARE THERE FUTURE SOLUTIONS?

Data has become a form of capital. Control of data as capital is and will continue to be a function of size.

1. Untested returns of new technologies may impact effectiveness of AI for small business. BD/AI is a relatively newly developing field, just as computing was at the dawn of the personal computing era when expensive minicomputers dominated the business landscape. Then there was no need, and in fact every objection, to small businesses using computing before it was mature enough to provide actual economic value to enterprises. Supervised or unsupervised, today’s machine learning is fragile. Its results remain less certain than the hype. Big data is expensive to collect and its validity has been justly questioned. Validity is not established by the size of BD data banks, because of uncontrolled biases built into the questions used to generate the data in the first place.

2. Unreliability of data creates asymmetric costs for small business. The ways BD/AI use data tend to work in the direction of confirmation bias. That is, they reinforce racial, ethnic, size, and other prejudices. We see that in such disparate fields as intelligence testing and bank lending. It also is the reason that police departments are increasingly rejecting facial recognition in the “post Floyd” era. For small businesses with less data and less ability to recover from failure, even minor confirmation biases can engender disastrous impacts. While software upgrades and patches are being developed to make both AI and BD more reliable, they have not created the certitude small business need in order to invest with confidence.

3. Lack of time and resources impact competitiveness of small business. Time is a more restricted capital asset than money, because unlike money once lost it cannot be recaptured. The time to implement AI solutions and the scarcity of qualified practitioners is also the liability in the competition between small businesses and larger concerns. The Wall Street Journal reported in 2020 that 21% of small businesses use AI now or plan to do so within two years, compared to 65% reported by large companies. It does not support either statistical finding.

4. Return on investment may be too long for small business to absorb. In an era of rapid new software development businesses can either make or buy technology. Neither option is a good one for BD/AI. Building platforms is prohibitively expensive in time and money for most small businesses. While there are ready-made solutions on the market, small businesses on brittle budgets in highly competitive businesses need to consider carefully before adopting them. The efficacy and ROI of these technologies have not yet been firmly established in practice. AI for AI’s sake, or for promotional or branding purposes is a poor idea.

5. Utopian expectations distort the decision-making process for small business. Karen Mills, a Harvard Business School professor describes what she calls a “Small Business Utopia” in which AI powers and enhances accounting and management functions and leads to a stronger competitive positioning against large companies. But utopias tend to exist on an ideal, not a real plane. Small businesses typically use AI developed elsewhere and sold to them. That makes AI a kind of “black art” that can turn on its users.

6. Dominance of large companies in using AI is a serious and intended threat to small business. Purchasing AI generated data and analytics from Google, Amazon, and other large users of sophisticated AI programs for marketing and operational purposes may be a short term fix that creates a long term nightmare. They and others of almost equal size state that the sale of their BD/AI driven data is a strategic goal. They already have a near monopoly over data collection analysis, plus its selling and reselling. Each of these major purveyors of BD/AI data also has both strategic plans and a historic penchant for squashing startups and small companies if they pose a threat, especially in the retail space.

7. Those who make the case for AI in small business use skewed data. The analysts most enthusiastic about the benefits of BD/AI typically use only large company, or even multiple large company collaboratives to make their case. Consider, for example, the much-ballyhooed self-driving car. Grounded on BD/AI software that integrates and controls a vast collection of sensors and mechanical devices, only the Googles, Amazons, Teslas, Ubers Lyfts, and their ilk can play. While a small company can supply some components, the BD/AI function which commands the device will stay with the major companies. Consider also that all major self-driving vehicles are behind schedule in being deployed and making even a tiny profit. Even where small businesses can benefit from playing on this field, not many could stand the financial ravages brought on by delays in being paid.

8. Large businesses will resist the needed government response to help businesses adapt to the world of AI. Government regulation has the potential to level the playing field in data collection and analysis, and to protect the data of individuals and small companies. However, the prospect of them doing so is problematic. Large companies tend to influence such regulation as well as the interpretation of the rules after regulation is implemented. Large businesses always have done so. Personal data, the source of marketing BD, is under attack by Amazon, Google, and other social media companies. Despite some regulation in Europe, data rights have not been well defined and we have yet to see how the restrictions play out. It is simply not an arena that suits small players.

The Covid-19 epidemic, the ensuing economic crisis, and the crisis in racial relations bring us back a decade before the birth of artificial intelligence. Atomic energy, not AI, was the new and emerging technology in a Cold War world. How it would be used was not clear. In 1947, Vannevar Bush, a key player in the Manhattan Project that created the Atomic Bomb and now Head of the U.S. Office of Scientific and Technological Research, wrote that the new technology posed “a peril and a hope.” The tag line for AI and small business might read the same way.

* This is an abbreviated version of a more technical and annotated article, “Artificial Intelligence and Small Business: Its Past, Power, and Prospects,” Minority and Small Business Review, 2020, pp.5–14. Copy available from philawed@gmail.com upon request.