Kirk Borne’s five principles apply familiar defensive-driving habits to autonomous intelligent systems (AIS): look ahead, understand the whole situation, keep monitoring, preserve a safe alternative, and communicate so other participants can respond. They are a conceptual analogy for designing and supervising AI-enabled operations—not a published autonomous-systems standard or a safety method validated by outcome studies.
Borne presented the framework in Data Science Central on January 25, 2023; the page also shows November 30, 2024 in its date metadata. He describes AIS as an interdisciplinary field in which people interact with autonomous AI and argues that human intervention can still provide useful mid-course correction.
What are the five principles of safe driving in autonomous intelligent systems?
- Aim high in steering: reason about the wider route and likely future conditions, not only the system’s immediate state or past events.
- Get the big picture: maintain situational and social context, including how other actors are behaving and how circumstances may create opportunities.
- Keep your eyes moving: continuously monitor changing conditions and discard distractions that are interesting but operationally unimportant.
- Leave yourself an out: anticipate unexpected behavior and preserve a safe fallback or decision path.
- Make sure they see you: reduce risks caused by incorrect assumptions through connection, communication, collaboration, sharing, trust and empathy.
Borne says these rules are adapted from the Smith System of Safe Driving. That lineage supplies the analogy; it does not by itself establish that the principles improve AIS safety.
1. Aim high in steering
In driving, looking farther down the road gives a driver time to react. In AIS, the equivalent is to combine current observations with likely future conditions and the broader objective. A system should not optimize only for the next event if that choice creates a foreseeable problem later.
#1 Best Overall
Borne associates this principle with diagnostic, predictive and cognitive analytics. Diagnostics help explain what is happening, predictive methods estimate what may happen next, and cognitive analysis supports interpretation and judgment when the situation is ambiguous.
2. Get the big picture
An isolated data point can be correct yet misleading. “Big-picture” operation means placing observations in context: the operating environment, the goals of other participants, dependencies between systems and changes that can turn a constraint into an opportunity.
Borne links this idea to descriptive, prescriptive and cognitive analytics. Description establishes the current state, prescription evaluates possible actions, and cognitive capabilities help interpret context and human factors.
3. Keep your eyes moving
Autonomous systems need ongoing attention rather than a single successful check. Inputs, alerts, user behavior and external conditions can change after a decision has been made. Monitoring should therefore prioritize signals that can alter risk or the next action, while suppressing noise that merely attracts attention.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Borne connects the principle to diagnostic analytics and event alerts. In practice, an alert is useful only when its meaning, urgency and owner are clear enough to support a timely response.
4. Leave yourself an out
A robust AIS should avoid decisions that leave no safe recovery option. Before committing to an action, it should consider plausible deviations—such as another actor behaving unexpectedly, a sensor becoming unreliable or a downstream service failing—and retain a fallback, pause, escalation or alternative route where feasible.
Borne associates this principle with prescriptive analytics, which evaluates and recommends courses of action. The rule is a design mindset, not a specified redundancy level or a guarantee that a safe alternative will always exist.
5. Make sure they see you
Many failures occur because participants assume that someone else has detected an action, intention or hazard. AIS can reduce that risk by making its state and intended behavior understandable to relevant people and systems, and by supporting explicit coordination rather than silent assumptions.
Borne ties this principle to prescriptive and cognitive analytics and names connection, communication, collaboration, sharing, trust and empathy as parts of the relationship. In a human-facing deployment, that can include clear status messages, escalation channels and explanations appropriate to the decision’s stakes.
How the principles fit Borne’s self-driving-enterprise model
Borne describes a self-driving enterprise—including self-driving cars—as having five characteristics. He maps each characteristic to an analytics dimension:
| Self-driving-enterprise characteristic | Analytics dimension in Borne’s scheme | Relationship to safe-driving principles |
|---|---|---|
| Sensing and streaming | Descriptive analytics | Supplies an up-to-date view of what is happening. |
| Responsiveness | Diagnostic analytics | Helps identify significant changes and causes quickly. |
| Learning and agility | Predictive analytics | Supports anticipation of future states and adaptation. |
| Contextual optimization | Prescriptive analytics | Balances alternatives and retains safer choices. |
| Deciding and acting | Cognitive analytics | Connects interpretation, judgment and action in context. |
This table is Borne’s explanatory mapping. It should not be read as a tested safety protocol, certification checklist or evidence that one analytics category guarantees a particular capability.
Why human oversight remains part of the model
Although “autonomous” can imply operation without people, Borne deliberately leaves room for intervention. His stated view is: “While full autonomy suggests that the system can operate without human interaction, it is useful to leave open the opportunity (or even the essential necessity) for human intervention, to provide mid-course corrections that keep the AIS on the right (and ethical) course.”
Recommended Free Tools
That position treats autonomy as compatible with supervision, escalation and correction. The appropriate intervention design will depend on the system’s risk, timing and authority; the five principles do not specify staffing, response-time targets or an approval workflow.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What these principles do—and do not—establish
What they contribute
- A memorable way to organize forward-looking planning, context awareness, continuous monitoring, fallback planning and communication.
- A bridge between defensive-driving language and analytics functions used in autonomous or highly automated operations.
- An explicit reminder that human correction and ethical oversight may remain necessary even when a system is described as fully autonomous.
What is not established by the source
- No quantitative safety improvement, benchmark, incident reduction or other outcome statistic is reported for the five principles.
- No empirical evaluation demonstrates that the analogy works across vehicles, enterprise workflows or other AIS deployments.
- No implementation requirements define sensors, alert thresholds, redundancy, explainability, governance or acceptable risk.
- No competing five-principle AIS framework is established for a direct comparison.
How to use the framework responsibly
- Define the operating context: identify the system’s decisions, affected people and external actors before translating a principle into a control.
- Make future risks explicit: document plausible near-term and downstream conditions rather than optimizing only the current state.
- Set monitoring priorities: distinguish safety-relevant events from low-value alerts and assign clear response ownership.
- Design fallback paths: specify when the system pauses, switches modes, asks for help or hands control to a person.
- Make behavior visible: provide communication and coordination mechanisms suited to both human users and connected systems.
- Test the resulting controls independently: treat the five principles as design prompts, then validate concrete requirements with domain-appropriate testing, safety analysis and governance.
Bottom line
Borne’s five principles offer a practical vocabulary for thinking about safer behavior in autonomous intelligent systems: anticipate, contextualize, monitor, preserve options and communicate. Their value is explanatory and design-oriented. The source presents an adaptation of the Smith System, not an empirically validated AIS standard, so any real deployment still needs its own measurable requirements, testing and human-oversight plan.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
