Over a million developers have joined DZone.
{{announcement.body}}
{{announcement.title}}

What Prevents Companies From Realizing the Benefits of AI?

DZone's Guide to

What Prevents Companies From Realizing the Benefits of AI?

Here's what 22 executives who are familiar with AI and all its variants said when we asked them, "What are the most common issues you see preventing companies from realizing the benefits of AI?"

· AI Zone ·
Free Resource

Start coding something amazing with the IBM library of open source AI code patterns.  Content provided by IBM.

To gather insights on the state of artificial intelligence (AI) and all its variants — machine learning (ML), deep learning (DL), natural language processing (NLP), predictive analytics, and multiple neural networks — we spoke with 22 executives who are familiar with AI.

We asked them, "What are the most common issues you see preventing companies from realizing the benefits of AI?"

Here's what they told us:

Lack of Understanding

  • We have an AI platform and SaaS applications enabling clients to make apps smarter with AI but companies don’t know where to start. Start with a simple proof of concept with measurable and actionable results.
  • They are not able to see the capabilities of how AI can improve their business processes. Once you’ve helped them see the possibilities, you need to assess the entire organization prioritizing how AI can benefit different areas. Finally, how to integrate AI into existing systems and processes. AI cannot just say trust me. It must explain itself with powerful visualization. Just operating in a black box is not acceptable.
  • Medium to large enterprises have a mix of marketing needs for blue tooth low energy and IT for WiFi. Organizations are not prepared to manage across those two siloed domains. There’s a new set of talent in the IT department from configuration to the open source programmer who moves data around.
  • Trying to do too much. Realize the power of data and AI, hire a lot of people and solve big challenges. Don’t try to do too much — start small and scale.
  • You need enthusiasm for AI/ML/analytics among users and a top down mandate. The combined ability to encourage self-service analytics with governance. IT provides governance and support, single sign-on, management access, query optimization, data mashup. A three-legged stool with 1) IT/data management, 2) analytics, and 3) business analysts all supporting results in more successful adoption.
  • 1) Look at replacing humans completely versus focusing on using the domain knowledge and skills. Ability to augment human insights and creativity. Shared goal of the business model and determine the tools and solutions to use. 2) Start with the technology first for business process transformation. What are you trying to accomplish?
  • AI is still in the early stage of mass adaptation, so organizations are just starting to realize the cost benefits and value AI can provide. Most companies have large complex infrastructure that is not easily scalable for AI integration.
  • The most common issues we see are around lack of vision, understanding what AI truly is — hype vs. reality. The budget of course. But also fear of emerging technology.
  • Unclear business objectives. Trying to do too much. Data issues. Lack of expertise. Unreasonable expectations.

Lack of Expertise

  • Lack of focus, lack of data, lack of people with the right frame of mind, numeracy, and practicality. An empirical attitude is required to train technically. Be able to interpret results analytically. Be close enough to see data politics in action. Maintain control over quality. Have empathy for why people are trying to sabotage your ML process.
  • Pods of three people are collaborating to solve AI/ML problems: 1) database administrator (DBA) is pulling the correct data; 2) business intelligence (BI) analyst is identifying the business problem to be solved; 3) the deep learning (DL) specialist is creating the model. We provide an Apple-like solution to run machine learning while operationalizing SQL. Companies may have multiple pods of these three people trying to find solutions with neural networks, needing feature sets and training data to make accurate models. Tremendous skill sets are required. We are trying to bring SQL back to let the experts create libraries — ever richer and ever improving. Provide the ability to launch user-defined functions with pure SQL. Less sophisticated programmers are not experts in Python or neural networks. The right collection of technologies to work with each other like Hadoop and Spark which can be difficult to learn and expensive to implement. Fewer transfers of data and faster in software architecture.
  • The talent gap and familiarity with the underlying technology. AI has been around a long time but there are only a few experts. The field is growing quickly. Given the shortage of talent, there is massive competition. Enable developers to collaborate with data scientists. The other issue is that this is an entirely different way of programming. It is the inverse of what normal programmers are used to doing. With AI/ML you get data with output and set up the system that will learn what function is. This is a completely different mindset than developers have been brought up with.
  • While most fields within AI and traditional machine learning are easily accessible for most companies, the most advanced subfield of AI is deep learning, and most of the substantial improvements in recent years have been achieved using this specific sub-field of machine learning. Unlike other fields within AI, the barrier of entry for deep learning remains extremely high, and deep learning scientists are highly sought after by companies such as Google, Facebook, Microsoft, Apple, etc. As a result, most other companies are struggling in obtaining the required expertise for the application of deep learning.

Lack of Data

  • Get all of your data together, with meta data, into a data lake. You may need to make an investment up front. However, over time the benefits of AI far outweigh the cost. Once you make the upfront investment you’ll see cost savings and increased revenue quickly. Lack of knowledge of ML. Willingness to explore; however, no guarantee of positive results. You are guaranteed to learn.
  • The biggest opportunity that I see under all of these AI solutions is data. The biggest challenge right now, once you’ve built the AI solution, is allowing it to learn properly. Just like a human, if you teach it the wrong skills, it’s going to give you the wrong outcomes or behavior. Right now, most of these tools or bots require highly structured data. Currently, they aren’t advanced enough to learn off of unstructured data, such as Wikipedia, like humans are. Therefore, I think the biggest opportunity is to create bots or machine learning algorithms that can be applied to unstructured data — that’s the next frontier.
  • Data is all over the place and is difficult to wrangle. Microsoft is bringing an IoT platform to bear. Hortonworks and Cloudera are bringing data to a single repository.
  • Used to be constrained by the testing data but this has been relieved by different techniques. We’ve learned the need to train on actual data and to learn iteratively training the model as we go. Able to do much quicker than in the past.

Lack of Trust

  • We work with operations technology and explain things to them in ways they can understand. We build the tools to help them achieve the outcomes they are looking for. Brownfield equipment owners can be very uncomfortable with anyone interfering with their very expensive and precise equipment. We don’t need to touch existing systems. We work with what clients have and put the software in place to solve the problem at hand. Once the client sees the results of your use case you get access to their equipment. If you detect failure you can make adjustments to the equipment. Once a client becomes comfortable with what you can accomplish, they are open to automation. Autonomous cars with Bosch gateways for obstacle detections, field optimization, and accident avoidance.
  • Not having the right skillset. We automate the process so data scientists are not as necessary. Human nature and the resistance from end users to trust the data and resistance to change. Once end users see that the data is more accurate than their experience, they begin to trust the data.

What are the most common issues you see preventing companies from realizing the benefits of AI?

Here’s who we talked to:

  • Gaurav Banga, CEO, and Dr. Vinay Sridhara, CTO, Balbix
  • Abhinav Sharma, Digital Servicing Group Lead, Barclaycard US
  • Pedro Arellano, VP Product Strategy, Birst
  • Matt Jackson, VP and National General Manager, BlueMetal
  • Mark Hammond, CEO, Bonsai
  • Ashok Reddy, General Manager, Mainframe, CA Technologies
  • Sundeep Sanghavi, Co-founder and CEO, DataRPM, a Progress Company
  • Eli David, Co-Founder and Chief Technology Officer, Deep Instinct
  • Ali Din, GM and CMO, and Mark Millar, Director of Research and Development, dinCloud
  • Sastry Malladi, CTO, FogHorn Systems
  • Flavio Villanustre, VP Technology LexisNexis Risk Solutions, HPCC Systems
  • Rob High, CTO Watson, IBM
  • Jan Van Hoecke, CTO, iManage
  • Eldar Sadikov, CEO and Co-founder, Jetlore
  • Amit Vij, CEO and Co-Founder, Kinetica
  • Ted Dunning, PhD., Chief Application Architect, MapR
  • Bob Friday, CTO and Co-founder, and Jeff Aaron, VP of Marketing, Mist
  • Sri Ramanathan, Group VP AI Bots and Mobile, Oracle
  • Scott Parker, Senior Product Marketing Manager, Sinequa
  • Michael O’Connell, Chief Analytics Officer, TIBCO

Start coding something amazing with the IBM library of open source AI code patterns.  Content provided by IBM.

Topics:
ai ,machine learning ,deep learning

Opinions expressed by DZone contributors are their own.

{{ parent.title || parent.header.title}}

{{ parent.tldr }}

{{ parent.urlSource.name }}