Introduction to AI
Artificial Intelligence (AI) has been a topic of fascination and speculation for decades. This article explores the historical development, current applications, and future prospects of AI.
Historical Background
The concept of artificial intelligence was first introduced in the 1950s by John McCarthy, who coined the term. Since then, AI has evolved significantly with advancements in computing power and data availability.
Key Milestones
- - 1956: The Dartmouth Conference
- 1980s: Expert Systems
- 2010s: Big Data and Machine Learning
- 2020s: Deep Learning and AI Ethics
Current Applications of AI
Today, AI is integrated into various sectors including healthcare, finance, transportation, and more. Some notable applications include natural language processing (NLP), computer vision, and autonomous vehicles.
Why Companies Need To Institute Stricter AI Policies
As the hype grows around AI and more people try it out, many are using it at work to see how it can help them become more efficient. But there’s a real problem: Some employees are using public instances of chatbots, putting proprietary company data at risk by giving an open LLM access. Others use the company’s access to AI, but input highly sensitive and personal data, like Social Security numbers and financial data.
A new study from technology security company Kiteworks found that 27% of companies reported that nearly a third of all of the data sent to AI systems is the type of information that should be kept private, like company records, employee information and trade secrets. (It could be more; 17% of companies don’t know how much private data ends up getting sent to AI.) It’s a problem that’s growing. A Stanford University report on AI found a 56.4% increase in security incidents with the technology last year.
While it may seem obvious to the tech savvy, many employees might not know the risks of this kind of AI sharing, and 83% of companies only rely on training or warning emails to let them know. Kiteworks found that just 17% have automatic controls that keep employees from uploading sensitive information to public AI tools. Further, most companies don’t have much of an AI governance structure—only 9%, according to Deloitte research cited by Kiteworks.
The study results show that companies need to add policies and infrastructure to control employee use of AI and protect their own data. This kind of use can cause real damage to companies—not to mention their employees and clients. The study concludes that companies need to acknowledge the threat, deploy controls that can be verified, and ensure that they can stand up to regulatory scrutiny.
"With incidents surging, zero-day attacks targeting the security infrastructure itself, and the vast majority lacking real visibility or control, the window for implementing meaningful protections is rapidly closing," Kiteworks CMO Tim Freestone said in a statement.
How healthcare facilities can prepare their data for AI-assisted contract management
Until recently, healthcare, pharmaceutical, and medical-technology companies have focused the bulk of their [AI investments on clinical uses](https://www.businessinsider.com/pharmaceutical-companies-embrace-ai-in-drug-discovery-efforts-2025-3). AI has been employed to help diagnose breast cancer, identify potential patients for clinical trials, and accelerate the processes used to determine whether a device or a new drug has the potential to work or fail.
Now, medical centers are investing in AI for backend uses, such as speeding up contract generation, management, and review.
Prepare data for effective and accurate algorithm input
Creating uniform data sets is an essential step for any organization seeking to build a [large language model](https://www.businessinsider.com/chatgpt-openai-ai-dictionary-help-understand-nvidia-2023-12) or deploy an effective chatbot, Gould said.
He said that the data that's used to train any given LLM has to be accurately classified or organized into a database.
Structured data, by its nature, is classified. This includes information that's already in a healthcare system's database, like customer ID numbers, diagnostic codes, and prices for supplies.
But unstructured data — like a trove of PDFs outlining contracts, for instance — also needs to be properly prepared so that an AI algorithm knows which information to extract and what to do with it.
Because contracts from different vendors aren't written with a uniform structure, a person has to train [machine-learning AI](https://www.businessinsider.com/guides/tech/what-is-machine-learning) to look for specific things if they want to extract necessary data from a specific table, for instance, said Gould.
Imagine a PDF that's a contract. AI software 'can only tell you that it's a document — it doesn't tell you what that context is, whether it's a contract, an amendment, or a notice,' Gould told BI.
The algorithm only works well if the data matches what the algorithm is looking for. Saying, 'OK, chatbot, look all across my enterprise for information and see what you can find' is not effective, because the processing cost is going to become very, very expensive. And that's where you get drift and bias, because the data is not really ready.
Preparing data may also mean reckoning with historical information that wasn't properly classified or recorded, and ensuring that it's updated with the correct metadata for AI-based analysis. That can be time-consuming, Gould said.
First, data needs to be classified correctly and also within the appropriate context. That might mean identifying and fixing documents' previously mistaken tags. The data also needs to be stored in a way that maintains compliance with privacy and other safety regulations, like [HIPAA](https://www.businessinsider.com/guides/health/what-is-hipaa), which protects patient privacy.
Consider the impacts on personnel
Automating laborious manual tasks can be transformative, particularly for roles in procurement and compliance, said Matt Parker and Jacob Thompson of SpendMend, an AI tool for pharmacy procurement, at a recent webinar on AI integration.
In the world of pharma and healthcare, the people that are being asked to do this work are highly educated, expensive, and ambitious,
Still, some employees may need retraining or additional training because AI often changes job functions, Jeremy Strong, the vice president of supply chain at Rush University Medical Center, told BI. Ensuring there's a plan in place to address [AI upskilling](https://www.businessinsider.com/experiment-with-ai-tools-to-upskill-employees-says-google-researcher-2025-5) and acknowledging the significance of changes to employees' job functions can help to manage the transition, Strong said.
The better employees get at asking precise questions, the more AI algorithms can improve at providing accurate answers, said Gould. He gave the example of figuring out how many contracts with a specific type of clause will expire in 30 days. Typically, this process would take weeks, if not months, with a records manager or an entire records department looking through thousands of contracts. But with AI, an employee can learn how to ask a powerful and precise question that captures this information more quickly.