Responsible Use of AI with Deployed
How we write anything in business is has completely changed in 2023. It’s a shift as fundamental as the creation of the internet itself, the polar opposite of this poorly aged prediction 📠 = 💻.
Of all important business documents, writing a scope of work and then a contract for those services is perfectly positioned for AI. It’s creative (so great for generative AI and assisted writing) and it’s also contractual (so great for the guardrails from machine learning models that look for risky wording).
Deployed’s approach to AI is to make it as simple and effective as possible for every author and editor– using AI with pre-written prompts embedded in every text box to help you write, edit and publish the best written briefs, scopes and statements of work.
Is email riskier than AI? Probably.
We address how we approach the risks of AI below, but it is also important to highlight the risksthat exist today in drafting documents. Most company's existing scope, pricing and contracting process is a patchwork of Word, Excel and PowerPoints sent amongst internal users and their providers.
This brings its own significant risks; emailing very important financial and strategy documents isan enormous risk to data – and they reside on company, providers and users’ desktops and download folders for years to come with limited to nearly no traceability.
We believe that using controlled and auditable AI in a collaborative platform where there is no abilityto email outside the system, and where Personally Identifiable Information (PII) is constantly tracked, is less risky than uncontrolled authorship and email sharing that happens via existing legacy processes.
Artificial Intelligence and Machine Learning
First about AI and machine learning - while AI and ML are not quite the same thing, they are closely connected. The simplest way to understand how AI and ML relate to eachother is:
- AI is the broader concept of enabling a machine or system to sense, reason, act, or adapt like a human
- ML is an application of AI that allows machines to extract knowledge from data and learn from it autonomously
You can read more here as a helpful guide. For the purposes on this document, we will simply use the phrase AI, even though many of the applications of Deployed are ML.
Use of AI in freetext boxes
Where, specifically, is Deployed using AI in the platform? This is important to set context for the use of AI in each client’s use case. We use them in free text boxes.
Deployed has engineered AI into each free text box that requestors type into. Each question is pre-engineered. What does this mean?
A question like: “what is the scope of the work?” has prompts engineered so when the requestor clicks on the ‘✨’ button the prompts could be (i) give an example of a scope of work with clear deliverables (ii) simplify this scope of work to be consistent with the client’s tone of voice’ or (iii) elaborate on the response to provide three examples of other similar or adjacent services that may be requested for this scope.
A question like: “what problem are you trying to solve?” has prompts engineered so when the requestor clicks on the ‘✨’ button the prompts could be (i) give an example of a problem statement that is clear about the problem trying to be solved and write it in simple terms that anyone could understand (ii) simplify this problem statement to be consistent with the clients tone of voice or (iii) take the problem statement as drafted and provide more context and usea 5 why approach to break down the content.
The concept here isto ensure that each question is as easy as possible to answer and the author does not have to interact or prompt the AI, as it is already completed in the template maker.
As you can see here, the prompts are not taking the content of the text box to compare it to other clients, or previous work or outside documents. It is enhancing each responseto make it easier to understand, and therefore more likely for the wording to be clear as a result.
Responsible use of AI
This Responsible AI Policy is designed to address your concerns regarding Privacy, Confidentiality, Accuracy, Deployment, Compliance and Intellectual Property specifically within the context of Artificial Intelligence. We are committed to maintaining rigorous standards that protect your data and ensure the responsible use of AI technologies.
Privacy Considerations in AI
Data Minimization: Our AI systems are designed to require the least amount of personal data. We ensure that any data used solely serves the functional purpose of the AI application.
Encryption Protocols: All data ingested by our AI models is stored and processed with high-grade encryption to protect against unauthorized AI data manipulation.
Informed Consent: Prior to interacting with our AI services, users are made aware of the specific data types that the AI will process and for what explicit purpose.
Exclusivity: We commit to not using client-specific data to train orfine-tune our AI models, thereby preserving the confidentiality of your unique data.
Confidentiality and AI
Selective Access: Access to data processed by AI is controlled via role-based permissions. Only designated personnel who are involved in AI oversight have access.
Non-disclosure Commitment: Team members engaged in AI development and management are under strict NDAs that extend to AI-specific data handling
Data Retention Policy: AI-processed data is retained only as long as required for service provision and AI model evaluation, and then securely deleted.
Ensuring AI Accuracy
Data Verification: Prior to its use in AI processing, data undergoes stringent quality checks to minimize model errors and improve reliability. This is done through testing multiple prompts for each question designed within the platform.
Ongoing Validation: Our AI models are subject to periodic audits and quality assessments to ensure consistent accuracy over time.
User Identity Confirmation in AI: Multi-factor authentication is employed to confirm that data fed to the AI for personalized services is only accessible by the authorized user.
Safeguarding AI Deployment
Model Robustness: Algorithms are tested against data anomalies andoutliers to ensure that the AI behaves as expected under a variety of conditions.
AI Audit Trails: All AI decisions, data transactions, and userinteractions are logged in a tamper-proof manner for accountability and traceability.
AI and Legal Compliance
Legal Adherence: We comply with data protection regulations that have specific provisions related to AI use, ensuring both general and AI-specific legal conformity
Transparency Reports: Our updates focus on AI data usage, including how AI algorithms make decisions, and what steps are taken to ensure these decisions meet ethical guidelines.
Safeguarding Intellectual Property in AI
Client IP Protection: Any intellectual property you bring into theinteraction with our AI services remains yours. We have safeguards to prevent unauthorized access or use.
Respect for Third-Party IP: We operate with integrity and due diligenceto ensure that Microsoft and Open AI services do not infringe upon the intellectual property rights of other entities. Their algorithms are either proprietary or licensed legally.
Non-use of Client IP: Your intellectual property will not be used to refine our models or for any other purpose beyond the specified service offering to you under your exclusivetenant of the service.
Legal Assurances: Contracts explicitly outline our commitment to safeguarding client and third-party intellectual property in relation to our AI services.
Our commitment to safeguarding user interests in the age of AI is ongoing and central to ourmission