Research Proposal

Data Collection and Methods
In order to identify whether availability is the most significant factor holding AI back from widespread adoption and whether AI as a service is the solution to these factors, qualitative research was conducted through interviews to learn the opinions of experts. Interviews were conducted with AI experts and Ph.D. students. Questions were asked about how current AI technologies are being utilized, the weaknesses and strengths of AI, why the technology is not being used more, and how people can make sure this technology is used in more applications.

Interviewee Date/Method of Interview
Shimei Pan, PhD Dec 1, 1:30 PM, ZOOM
Lujie Karen Chen, PhD Dec 4 2020, 3:30 PM, ZOOM
Jianwu Wang, PhD Dec 6 2020, 1 PM, ZOOM
Md Osman Gani, PhD Dec 8 2020, 3 PM, ZOOM


Results and Analysis

Figure 1
Dr. Pan Dr. Chen Dr. Gani Dr. Wang
How widespread is the use of AI and ML today? - AI use is widespread
- Impacts everyone on a daily basis (online websites)
- 30% utilization of AI in Education
- 50% AI Utilization in Healthcare
- Online education and healthcare tech allow for more data collection which can increase AI use.
- Computing power and new methods increase use.
- A lot of tasks to automate left.
- Everyone tries to use AI for their application.
- AI at an early stage. Long way to go.
Figure 1 demonstrates that all of the people interviewed agree that AI is being utilized in many ways. But 3 of the 4 interviewees indicated that there is room for improvement and development and the other simply stated that AI does not solve all problems. These results show that there is more to AI than costs and availability.

Figure 2
Dr. Pan Dr. Chen Dr. Gani Dr. Wang
AI Strengths - Data analytics. - Perception including vision and voice recognition. - Big data/finding relationships.
AI Weaknesses - There are many. - Planning, decision making, emotion, creativity, and collaboration.
- AI does not carry over experience from task to task.
- Ethics, biases, black box, causation, transparency. - Reasoning and data quality.
- Black Box transparency issue.
- AI can not explain or teach its findings to humans.
Figure 2 demonstrates that all interviewees indicated that current AI is good at analyzing data and finding relationships and one interviewee indicated that this allows for computer perception. All interviewees indicated that current AI has many weaknesses. Two interviewees indicated that there is a black box issue and a lack of understanding of causality in AI. One interviewee indicated that these weaknesses hold AI back from successful planning, decision making, emotion, creativity, collaboration, and carrying over experience.

Figure 3
Dr. Pan Dr. Chen Dr. Gani Dr. Wang
Effects of widespread Adoption - Widespread adoption inevitable
- Large impact on jobs
- AI will assist professionals
- People will seek more and new entertainment
- Human machine collaboration
- Many jobs like truck driving will be lost.
- Jobs will change, doctors can use AI to interpret x-rays, etc.
What is stopping Widespread Adoption? - Current AI does not solve every problem - Maturity of technology, costs of implementation, and social impact.
- Lack of trust in AI, trasnparency issues. - More time and development needed.
Figure 3 demonstrates that all interviewees believe AI will be widely adopted and peoples lives will change as AI either takes jobs or assists professionals with their jobs. 3 of the 4 interviewees believe that current AI cannot solve all problems and more time is needed for the technology to mature. And 2 interviewees indicated that there is a black box issue with AI limiting trust, and that the social impact of implementing this technology must be extensively considered.

Figure 4
Dr. Pan Dr. Chen Dr. Gani Dr. Wang
Can small businesses afford AI? - Technology is getting cheaper
- New models require more and more power however, are very powerful
- Worrisome trend appearing
- If the technology is adopted more and more the price will go down
- Software development costs will go up
-Hard to access technologies are now affordable and accessible through APIs.
- And as competition increases the prices go lower.
How are costs and availability changing and why? -AI models with big models and big data are very expensive and not sustainable.
- There are cheaper options. And HR is the largest expense of small businesses.
- Depends greatly on needs and circumstances.
-Off the shelf options are making it possible. -Technology is used more now in business than before. Investments are necessary and when the return is profitable businesses will take the leap.
-The technology has to and will mature and lower in cost.
Figure 4 indicates that all interviewees believe that some sort of AI is accessible to small businesses and individuals however they believe that costs are still high and AI cannot do everything right now. All interviewees indicate that the technology is becoming cheaper and as it is adopted more and more it will continue to become cheaper. 2 interviewees do note that further investment into new technologies will increase prices of new technology, and 1 interviewee is worried about this trend and believes that it is not sustainable.

Figure 5
Dr. Pan Dr. Chen Dr. Gani Dr. Wang
Effect of AI as a Service? - Will be widely used
- Can not solve all problems
- Some people would rather develop their own
- Will have a large impact on the adoption of AI
- Can not solve all problems
- Experts needed to customize and apply the AI
- For general purposes the tech will be used, but for specific tasks it is not great
- May be trust issues
- Most companies will likely use AI as a service
- Every company has different needs so the implementation is not that straight forward
- Development from scratch is not wise
What applications require custom solutions? - Security applications
- Sharing proprietary data with competitors
- Applications off the shelf solutions can not solve
- Depends greatly on the needs of the business and the given application.
- Off the shelf solutions may not suffice
- Customization and training may be necessary
Figure 5 indicates that all the Interviewees believe AI as a Service will have a large impact on the adoption of AI, and most people and companies will use one form of AI as a service. And all interviewees indicated that the needs of businesses and individuals may not be provided by AI as a service, and even if they are they will need a lot of customization and implementation. One interviewee indicated that one reason people may not use AI as a service may be due to security reasons as a company may not want to share their data with others. For that reason many larger companies will either assemble their own AI groups or invest in/acquire private groups researching and developing AI. Another solution to availability are application source interfaces which allows different apps and programs to communicate with each other and open source code. Dr. Chen said that people can use open source software and modularize parts of the technology. Tasks like object recognition can be modularized but someone has to put the parts together, train the system, and tweak it for the applications. Dr. Gani suggested that sensing APIs can be used and Dr. Wang said robotics technologies can be implemented with assistants like google assistant or alexa. Essentially developers and researchers can use existing technologies to build their own technology. As Dr. Wang put it, developing everything yourself is not wise.

Discussion/Conclusion

The research partially supported the hypothesis. As seen in figure 4, while cost and availability are major factors holding back the adoption of AI and AI as a service will have a great impact on the adoption of AI (Figure 5) as my hypothesis stated, AI as a service still requires experts and a lot of resources. And for many applications AI as a Service will not suffice and many might not want to share their data. Beyond availability and cost there are many limitations of AI. As seen in figure 2, AI has a transparency issue, a transparency/ethics/bias issue, a lack of understanding of causality, deficiencies in planning, decision making, emotion, creativity, collaboration, and the lack of the ability to carry over experience from one task to another. Similarly, as shown in figure 3 some of the leading factors holding AI back are that current AI cannot solve all problems, there is a lack of trust, and there is a lack of maturity requiring more time and development. The literature review pointed to the issue of transparency, lack of efficiency, lack of availability but not the limited capability like the lack of understanding of causality and the ability to carry over experience, which are massive roadblocks in the future of AI.

This study was only conducted on four AI experts all who work at the University of Maryland Baltimore County. Researchers at other institutions may have different opinions. This study and research can be redone with a greater and more diverse audience. The issues of transparency, the black box issue, public trust in AI, bias in AI, AI understanding of causality, AI transferring experience from one task to another, and measuring efficiency, and making AI more available and easier and cheaper for individuals and small organizations are all topics that can and should be further investigated.