Log Out

The Why and How of Artificial Intelligence

Artificial intelligence has matured from technology buried in computer science labs using complex coding techniques to more common algorithms and supporting technologies. 

By Randy Johnston 

One of the hot areas of emerging technology is artificial intelligence (AI). You know a topic is hot when products use the terminology in their name or in their marketing materials. We’ve seen this done with “ease of use,” cloud computing and blockchain. Artificial intelligence is so hot of a topic among the development community that marketing teams are saying products have AI when, in fact, they do not. Because of this, I’ve been using the approach of calling this “artificial” artificial intelligence or “fake” AI. Facts matter, and many products that claim to have AI simply do not. It is truly buyer beware right now in this area. 

As a computer scientist by training, I admire products that have developed solutions that leverage AI. Program development in this area is not easy, and the products are just starting to work and do meaningful tasks. Some of the goals of AI are quite lofty, and the promises and risks of AI in computing are quite large. Consider the following: 

On the positive side: 

  • Machines mimic cognitive functions associated with human minds, such as learning and problem solving. 
  • As AI becomes more capable, tasks that were considered AI are considered solved (e.g., optical character recognition). 
  • Today, AI developments include human speech, autonomous cars, interpreting complex data like images and video and more. 
  • Algorithms can learn from data and provide insight and actionable items with minimal human intervention. 

On the down side: 

  • For difficult problems, algorithms require enormous computation. 
  • Consider this quote from Stephen Hawking: “The development of full artificial intelligence could spell the end of the human race. Once humans develop artificial intelligence, it will take off on its own and redesign itself at an ever-increasing rate. Humans, who are limited by slow biological evolution, couldn't compete and would be superseded.” Other critics include Bill Gates, Elon Musk and Peter Thiel. 
  • Devaluation of humanity. 
  • Decrease in demand for human labor. 
  • Artificial moral agents. 
  • Machine ethics. 
  • Malevolent and friendly AI. 
  • Machine consciousness. 
  • Robot rights. 
  • Superintelligence. 

AI has been under development for a long time, starting at Dartmouth in 1956. I was using the programming language Lisp in 1975 and this tool was broadly used until 1987, when it was replaced by SmallTalk/Medley. AI is routinely divided into sub-fields such as robotics or machine learning. Traditional goals include: reasoning, knowledge, planning, learning, natural language processing, perception and explainability. Tools include versions of search and mathematical optimization, neural networks and methods based on statistics, probability and economics. Stuart Shapiro divides AI research into three traditions, which he calls computational psychology, computational philosophy and computer science. Together the human-like behavior, mind and actions make up AI. 

We’ve seen the results of AI with public relations stunts when: 

Why? 

AI is a broad collection of topics and approaches. Because there are so many topics to cover, there are broad fields of study that each have a lot of depth. However, purists are after the last bullet in the list below, general intelligence. This is not around the corner, as computer scientists of the 1950s believed, but a decade or more into the future, even with the rapid progress being made today. There are a number of problems AI is trying to solve: 

  • Reasoning: AI has progressed using "sub-symbolic" problem solving; statistical approaches to AI mimic the human ability to guess faster and more accurately. 

  • Knowledge representation: A representation of "what exists" is an ontology: the set of objects, relations, concepts and properties formally described so that software agents can interpret them. 

  • Planning: Intelligent agents must be able to set goals and achieve them, modifying inputs as needed. 

  • Learning: The study of computer algorithms that improve automatically through experience. 

  • Natural language processing: The ability to read and understand human language. 

  • Perception: The ability to use input from sensors (such as cameras, microphones, tactile sensors, sonar and others) to deduce aspects of the world — think digital exhaust and the Internet of Things (IOT). 

  • Motion and manipulation: Robots need to handle tasks such as object manipulation and navigation, with sub-problems such as localizationmapping and motion planning

  • Social intelligence: Affective computing is the study and development of systems that can recognize, interpret, process and simulate human affects (=emotions), needed for two reasons: 

    • Being able to predict the actions of others, such as in self-driving vehicles.

    • Facilitating human–computer interaction by showing emotions. 

  • Creativity: Theoretical and/or practical generation of novel and useful outputs including music and art. 

  • General intelligence: Researchers think that their work will eventually be incorporated into a machine with artificial general intelligence, while a few believe that anthropomorphic features like artificial consciousness or an artificial brain may be required for such a project. 

How? 

So how do artificial intelligence approaches work? They use: 

  • Cybernetics and brain stimulation: There is a connection to neurology. 

  • Traditional symbolic AI: John Haugeland named these approaches to AI "good old fashioned AI" or "GOFAI,” exploring the possibility that human intelligence could be reduced to symbol manipulation. 

  • Cognitive simulation: Economist Herbert Simon and Allen Newell studied human problem-solving skills from psychological experiments resulting in the Soar architecture in the 1980s. 

  • Logic: In his laboratory at Stanford (SAIL), John McCarthy used formal logic and led to the Prolog language and the science of logic programming. 

  • Anti-logic or scruffy: Marvin Minsky and Seymour Papert found that solving difficult problems in vision and natural language processing required ad-hoc solutions. 

  • Knowledge: Edward Feigenbaum of Stanford developed knowledge-based expert systems in the 1970s. 

What does this mean to the practice of accounting and to accountants? We have several working examples available: 

  • Accounting: Zoho Zia (Zoho Intelligent Assistant) workflow analysis, best time to contact, automatically research and suggest data to complete client records, email sentiment analysis 

  • Auditing: Mindbridge 

  • Financial services: Kasisto, Moneystream 

  • Sales: Salesforce PredictionIO. 

  • Self-writing applications: Crane.ai 

We see a number of products that are running crude AI today and vendors that are pretty far along in their use of AI. Getting products that apply to small and medium businesses are more of a challenge, but we see efforts at Intuit, Xero, BQE Core, Citrix ShareFile, Thomson, CCH and most other products that apply to the CPA profession and to small and medium businesses. 

As development continues and artificial intelligence transitions from an emerging technology to mainstream, vendors will choose from many open source and proprietary suites that have AI capabilities or they will develop their own algorithms inside their products. Examples today include: 

  • TensorFlow 

  • Theano 

  • Torch 

  • A list of 15 tools which includes comments and some of the above can be found here

  • Another good list of AI tools can be found here

The best example of tools for accounting that are working today is: 

  • IBM Watson: This tool has been configured for analysis of tax, as well as analysis of core financial and operational data by KPMG. 

  • SAS: Visual Investigator cost $1 billion to develop. David Stewart, director of Financial Crimes & Compliance of SAS, suggests 50–70 percent of banks’ compliance spend is on anti-money laundering. Financial firms spend 4 percent now and will spend about 10 percent in 2022. 

  • Zoho Zia (Zoho Intelligent Assistant): Workflow analysis, best time to contact, automatically research and suggest data to complete client records, email sentiment analysis. 

Here’s a summary of what you need to know about artificial intelligence: 

 

Key Questions 

Key Answers 

Why is the new technology better? 

It is a method of data analysis that automates analytical model building. 

How can you do this today? 

AWSAzureGoogle Cloud AIIBM WatsonSAS 

What are the risks? 

Wrong data set, conclusion unguided 

Where/when can you use it? 

When data can answer a specific question 

How much does it cost? 

Can be up to $10,000 per hour, or free on open source 

When is it expected in the mainstream? 

Simple AI now, fake AI common in current promotions, usable AI 4–6 years 

What technology or service does it displace? 

Repetitive or analytical human labor 

Other resources 

CPAPACSO 

 

Artificial intelligence has matured from technology buried in computer science labs using complex coding techniques to more common algorithms and supporting technologies used as part of the design strategy of new generation products. Many of the developers have known of AI techniques for years but did not have a practical way to apply the algorithms because the computing overhead was too high, the sample of data was too small and the number of techniques that needed to be applied made the code too complex. With centralized computing in SaaS applications and cloud data centers, AI has become much more practical and accurate. 

Recommended Next Steps 

Watch for applications that claim they have artificial intelligence. Do the applications truly exhibit learning? Are they rules- or forms-based and limited? If so, they may not be AI. Do they improve over time? Does more data make them more accurate? Can the applications make new conclusions without additional programming? If so, they may be truly be AI. 

You need to filter out products based on rules, forms or pattern recognition that are programmed to recognize each specific form/task and make a decision based on recognizing the form or task. You want the system to accept inputs of all kinds, recognize new data, learn about the data and make conclusions that provide insight. Like the human race, it is hard to predict where AI will take the capabilities of machines and computers. As many of you have heard me say before, computing can be used for good or bad. I tend to look at the bright side of life, as we are reminded by Monty Python here. While Skynet is possible, and we want to listen to the cautionary predictions of Hawking, Gates, Musk, Thiel and others, on the future of artificial intelligence. 

 

Randy Johnston headshot

Randy Johnston is a shareholder in K2 Enterprises, LLC, a leading provider of CPE to state CPA societies. He also owns Network Management Group, Inc., a managed services provider that provides around-the-clock support from Boston to Honolulu. Concepts for this article were extracted from the Emerging Technology session produced as part of the 2018 K2 Technology Conferences and from Johnston’s own experience working with technology at various firms in the United States. Ask for help at NMGI by emailing [email protected] or call (620) 664-6000.