Join top executives in San Francisco on July 11-12, to hear how leaders are integrating and optimizing AI investments for success. Learn More
When it comes to artificial intelligence (AI) adoption, there is a growing gap between the haves and the have-nots.
According to IBM, the global AI adoption rate went up by 4 percentage points in 2022, reaching nearly 35%. However, the study also found that the gap in AI adoption between larger and smaller companies also grew significantly in the past year.
Today, larger companies are twice as likely to have actively deployed AI as a part of their business operations than smaller companies, which are more likely to be exploring or not pursuing AI at all due to development cost and scalability issues.
“SMEs [small- to medium-sized enterprises] are often plagued with the problem of scaling their operations on account of huge financial implications,” said Dipak Singh, head of data science at Indus Net Technologies. “They often don’t want to venture onto the path of AI because they are not sure about the outcome of their AI projects.”
Join us in San Francisco on July 11-12, where top executives will share how they have integrated and optimized AI investments for success and avoided common pitfalls.
AI projects generally take months to develop and mature, bringing a long gestation period and significant expenses. That’s where AI-as-a-service (AIaaS) comes in: It was born out of a desire to democratize AI for all while addressing the growing demand for AI, cognitive computing and large-scale adoption of cloud-based solutions.
The global AIaaS market is poised to exceed $41 billion by 2025, nearly five times the size of today’s market, according to NASSCOM. AIaaS is often built on big cloud providers including IBM, SAP SE, Google, AWS, Salesforce, Intel and Baidu, but there are also a handful of startups offering solutions. Most recently, California-based API startup Assembly AI launched an AI-powered API to convert audio or video to text. Offered as an AIaaS model, the API empowers developers by aiding in-model development for transcribing, understanding and analyzing the audio data.
AI for all
AIaaS, which gained traction beginning in 2020 during the COVID-19 pandemic, works like any other “as-a-service” business model. It is composed of multiple delivery models and offerings including, but not restricted to, off-the-shelf tools for development, testing and deployment, and scaling of AI/ML models; vertical services provisioning such as inference-as-a-service, annotation-as-a-service, and machine learning-as-a-service (MLaaS); and fully outsourced or managed service models.
While the definition can be blurry, AIaaS does enable users to harness the power of AI/ML without writing a single line of code and without needing any particular technical expertise.
“Leveraging AI as a service-based model could potentially be the right approach for those companies who are yet to understand the scope to which AI could be incorporated into their business,” said Tushar Bhatnagar, cofounder and CTO at vidBoard.ai and cofounder and CEO of AIaaS provider Alpha AI.
Because AIaaS doesn’t demand large initial investments or massive human resources and involves lower risk, businesses can stay focused on their core competencies while getting access to the capabilities of AI/ML, Bhatnagar said.
AIaaS enables SMEs to quickly use pretrained models via plug-and-play mechanisms at a nominal cost. Some of the areas where AIaaS is helping companies include bots and digital assistants, cognitive computing APIs, machine learning frameworks and data labeling.
There is no free lunch: AIaaS is no exception
AIaaS is a boon for SMEs, but the offering is not flawless.
For example, Bhatnagar said that companies need to be aware of issues that could stem from inaccuracy in data insights, algorithmic bias and the “black box” nature of AI.
Take algorithmic bias, for example. An AI model creates a series of instructions that a system has to follow to achieve a particular task, which is typically created by humans. However, if the algorithms are flawed, not optimized for edge cases, or biased, they will provide unfavorable and unreliable results and conditions. If not kept in check, an AI-as-a-service model could scale up quite fast. In the end, companies might find themselves looking for more complex solutions and customizations, which can be more expensive and necessitate hiring and training more specialized personnel.
“When it comes to inaccuracies in data insights, AI programs can only learn from the information we present to them,” said Bhatnagar. “One’s results may be wrong or highly disjointed if the data provided to the program is incomplete, unreliable or lacks structure. As a result, AI can only be as smart, useful, or innovative as the data you feed into it.”
Lastly, AIaaS often lacks any explainability. While a solution may consistently return accurate findings, it cannot explain how it arrived at that particular conclusion.
Tips for navigating the AIaaS landscape
The best SME use cases for AIaaS, said Singh, are those “odd and mundane” operational activities that can be automated using AIaaS and do not involve private and confidential data, such as data labeling and classification, or bots and digital assistance.
Singh also recommends doing thorough background research about the AIaaS provider to make sure company data will be in safe hands.
Finally, companies should also create usage, access and security protocols. For example, if AIaaS is used for bots and digital assistance, what kind of data should the company allow the system to store or share? To what extent can they use it or not use it? What are the rules and regulations for employees moving in and out of the platform? This knowledge needs to be clearly documented and shared with the different stakeholders within the organization, said Singh.
VentureBeat's mission is to be a digital town square for technical decision-makers to gain knowledge about transformative enterprise technology and transact. Discover our Briefings.